a review of surgical robots for spinal interventions - … multiple electronic databases were...

A review of surgical robots for spinal interventions

Alvaro Bertelsen*Javier MeloEmilio SánchezDiego Borro

Applied Mechanics DepartmentCEITand Tecnun, University of Navarra,San Sebastián,Spain

*Correspondence to: A. Bertelsen,Paseo Manuel Lardizábal 15, 20018,San Sebastián, Guipúzcoa, Spain.E-mail: [email protected]

Abstract

Background This study aimed to describe the state of the art in surgicalrobotics for spinal interventions, a challenging problem for which robots canprovide valuable assistance.

Methods Multiple electronic databases were searched for articles publishedduring the last 10 years (2002–2012). Results were refined by definedinclusion criteria.

Results A total of 18 different robots were found. Among them, five arecommercially available systems: one specifically designed for spinal surgery,one for percutaneous needle-based interventions, two are radiosurgicalsystems with reported applications on the spine and another is a commercialrobot which has been experimentally tested on spinal surgery. The remainingprojects are research prototypes which are still on validation stages.

Conclusions A comprehensive state of the art is presented, showing thatspinal robotic surgery is still at an early stage of development but with greatpotential for improvement. Copyright © 2012 John Wiley & Sons, Ltd.

Keywords robot; spine; surgery; surgical robots; review; screw insertion; needleinsertion

Introduction

This article aims to describe the state of the art in robotics for spinal surgery, ademanding medical field for which robots already provide valuable assistance,although there is still plenty of room for improvement. Spinal surgery requiresa very high level of precision, as it has considerable risks due to the criticalstructures that surround the spinal column: blood vessels are found close tothe vertebrae and nerves connected to the whole human body are rooted tothe spinal canal. Damage to any of these structures can produce considerableside-effects, ranging from pain to paralysis. These higher-than-normal require-ments make robots ideal candidates for surgical assistants, as they can achievesuperior levels of precision, are not affected by fatigue and can perform repetitivetasks without decreasing their performance.

The development of robots for spinal surgery dates back to 1992, althoughthe largest part of the research effort is found during the last 10 years,i.e. 2002–2012, which is the period covered by this study.

Nowadays, robots are used for a variety of surgical procedures, such as trans-pedicular fixation. This type of surgery, which has attracted a lot of attentionfrom robotic scientists, consists of the fixation of two or more vertebrae bymeansof screws inserted through the pedicles, which are later connected by metal bars.The screws and rods form a rigid structure which prevents relative motionbetween the vertebrae, hence preventing further damage to the one affected byfracture or herniation. Transpedicular fixation, apart from the dangers inherent

REVIEW ARTICLE

Accepted: 1 October 2012

Copyright © 2012 John Wiley & Sons, Ltd.

THE INTERNATIONAL JOURNAL OF MEDICAL ROBOTICS AND COMPUTER ASSISTED SURGERYInt J Med Robotics Comput Assist Surg 2012.Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/rcs.1469

to all spinal surgeries, has additional problems, such aspossible pedicle fracture, difficult visualization of the surgi-cal area and high exposure of the patient, surgeon andclinical staff to ionizing radiation if X-ray imaging is usedto guide the intervention. All the aforementioned problemscan be solved, or at least alleviated, by the use of surgicalrobotic assistants, which can help the surgeon to insertscrews in precisely selected positionswithminimal deviation.

Apart from transpedicular fixation, robots are now usedfor needle-based procedures, such as biopsies (1,2),vertebroplasties (3,4) and facet blocks (5). Roboticsystems are also used for tumour ablation (6) and, to alesser extent, for tumour resection (7). They also help toreduce radiation doses to surgeons, patients and clinicalstaff during interventions (8). Evenmore, ongoing researchaims to develop robotized endoscopes capable of explora-tion and surgery in the sub-arachnoid space, which is onlya few millimetres wide (9).

This article is divided into four sections: criteria for articleinclusions are given in Methods, while chosen projects aredescribed in Results; an analysis of the state of the art isgiven in Discussion, with final remarks in Conclusions.

Methods

Publications about surgical robots for the spine weresearched on the IEEE Xplore, PubMed, Google Scholar andCiteSeerX databases. On them, searches were performedfor articles including the terms ‘spine’, ‘vertebra’ or ‘pedicle’,but always with accompanied by the word ‘robot’. In addi-tion, the Medical Robotics Database (MERODA) hosted bythe University of Heidelberg1 was searched for the sameterms. Only the results which described mechatronicsystems which performed clearly defined surgical tasks inan autonomous or semi-autonomousmanner were included.

In particular, robotic systems designed for the followingapplications were discarded, as they fell out of the scopeof this article:

• Simulators for surgical training.• Testing of spinal specimens (e.g. load-bearing or kinematic

analysis) helped by robots.• General-purpose robots which, in theory, could be

deployed in spinal surgery but without reports ofactual experiments.

• Robotic systemswhich only performed image acquisitiontasks (e.g. cone-based computed tomography).

Results

Applying the methods described in the previous section, atotal of 18 robotic projects were found, which are summa-rized in Table 1. Descriptions of them are given in the

following sections, citing the applications for which theywere designed, their main technological features and asummary of their performance results.

Early systems (before 2002)

The earliest work on robotic-assisted spine surgery wastraced back to 1992 and was reported by a research teamfrom Grenoble, France (10). As in many other early studieson surgical robotics, the authors adapted an industrialrobot, in this case a PUMA 260, for use in the operatingroom. The robot was designed as an assistant for transpedi-cular fixation, holding a laser guide which pointed drillingtrajectories over the patient’s vertebrae. Surgical planningwas carried out on a segmented pre-operative computedtomography (CT) scan, which was registered to intra-operative X-ray images. The authors presented a drillingexperiment on plastic vertebrae, on which they claimed toobtain sub-millimetre accuracy.

In 1995, Santos-Munné et al. (11) proposed anotherrobotic system for transpedicular fixation, which sharedmany similarities with the project first presented inGrenoble: it also advocated the use of an industrial robot(a PUMA 560), intra-operative X-ray imaging system andplanning of drilling trajectories on pre-operative CT scans.Differing from previous work, the proposed system placeda drill guide on the robot’s end-effector, which was madeof radiolucent material with embedded metal spheres. Inthis way, it could be located on the intra-operative X-rayimages and registered to the coordinates of the plannedtrajectories. The authors neither reported about theproject’s implementation nor gave experimental results.

Some years later, researchers from the FraunhoferInstitute developed the Evolution 1 surgical robot, whichwas commercialized by Universal Robot Systems (URS)and deployed on multiple clinical institutions. AlthoughEvolution 1 was designed for neurosurgery, a researcheffort was made to extend its use to spinal interventionsunder the project named ’Robots and Manipulators forMedical Applications’ or RoMed (12). However, URS wentout of business some time later, forcing its former clientsto stop using their robot, due to the cessation of mainte-nance and technical support (13).

Robots for screw insertion

The majority of robotics projects developed between 2002and 2012 – more precisely 8 out of 18 – have focused onscrew insertion, a task required for surgeries such astranspedicular fixation and cervical body fusion. Detaileddescriptions of these robots are given in the followingsections and a summary of the most relevant experimentsusing them can be found in Table 2.

SpineAssist/RenaissanceA major breakthrough for spinal robotic surgery came in2003, when a team of Israeli researchers presented theMiniAture Robot for Surgical procedures, or MARS (14).

1http://www.umm.uni-heidelberg.de/apps/ortho/meroda/robdat.php

A. Bertelsen et al.

Copyright © 2012 John Wiley & Sons, Ltd. Int J Med Robotics Comput Assist Surg 2012.DOI: 10.1002/rcs

http://www.umm.uni-heidelberg.de/apps/ortho/meroda/robdat.php

http://www.umm.uni-heidelberg.de/apps/ortho/meroda/robdat.php

Table1.

Summaryof

themainap

plications

,fea

turesan

dreferenc

esof

thestud

iedrobo

ticsystem

s

Nam

e(ormainreferenc

e)App

lications

Commercial

availability

Ope

ratio

nmod

es*

DoF

Tracking

tech

nology

Referenc

es

SpineA

ssist

Tran

sped

icular

fixatio

n,verteb

roplasty,b

iops

y,GO-LIF

Yes(FDAclea

ranc

ean

dCE

mark)

Assistiv

e6

Fluo

roscop

y(8,14–

20)

SPINEB

OTv1

Tran

sped

icular

fixatio

nNo

Assistiv

e,au

tono

mou

s7

IRtracking

(21)

CoRA

Tran

sped

icular

fixatio

nNo

Assistiv

e,tele-ope

rated

6IR

tracking

(22)

SPINEB

OTv2

Tran

sped

icular

fixatio

nNo

Assistiv

e5

Bi-plana

rfluo

roscop

y(23)

VectorBo

t/Kine

med

icTran

sped

icular

fixatio

nNo

Assistiv

e7

IRtracking

(24,25

)Neu

roglide

Cervical

interbod

yfusion

No

Assistiv

e4

IRtracking

(26)

(Wan

g201

0)Laminectomy

No

Auton

omou

s2

Forcesensing

(31)

MINOSC

Sub-arachn

oidsp

aceexploration,

localized

electro-stim

ulation

No

Tele-ope

rated

5En

doscop

e(9)

Acu

Bot

Verteb

roplasty,b

iops

y,ne

rvebloc

ksYe

s(FDAclea

ranc

e)Assistiv

e,tele-ope

rated

6Fluo

roscop

y,intra-op

erativeCT

(1,5)

Inno

motion

Biop

syPa

used

(CEmark)**

Tele-ope

rated

5Intra-op

erativeMR,

intra-op

erativeCT

(2)

DLR

LWRIII

Biop

sy,v

ertebrop

lasty

No*

**Assistiv

e,tele-ope

rated

7IR

tracking

,3Dradiog

raph

y(3)

Cybe

rknife,N

ovalis

Tumou

rab

latio

nYe

s(FDAan

dCE

mark)

Auton

omou

s6

Fluo

roscop

y(6,43–

47)

daVinc

iTu

mou

rresection

Yes(FDAan

dCE

mark)

Tele-ope

rated

7(for

each

arm)

Endo

scop

e(7,40–

42)

RIME

Tran

sped

icular

fixatio

nNo

Tele-ope

rated

6IR

tracking

(27,28

)(O

nogi20

05)

Verteb

roplasty

No

Auton

omou

s5

Fluo

roscop

y,IR

tracking

(4,36)

SpineN

avVe

rteb

roplasty

No

Auton

omou

s,tele-ope

rated

5Intra-op

erativeCT

(37,38

)RS

SSTran

sped

icular

fixatio

nNo

Assistiv

e5

IRtracking

(29,30

)

*Assistiv

e,therobo

tacts

asamereassistan

t,ke

epingthesurgical

toolsin

placewhile

thesurgeo

nman

ually

performstheactual

drillingor

insertion;

Auton

omou

s,therobo

tiscapa

bleor

drillingor

instrumen

tinsertionby

itself,with

outdirect

interven

tionof

thesurgeo

n;Te

le-ope

rated,

therobo

treplicates

thesurgeo

n’smotion,

althou

ghthelatter

controlsitremotelyfrom

aco

nsole.

**Th

eInno

motionco

mmercializationwas

stop

pedin

2010

,alth

ough

itisexpe

cted

tobe

restartedin

2012

.***The

system

prop

osed

byTo

var-Arriaga

(3)isaresearch

prototyp

e,althou

ghitisba

sedon

theLW

Rrobo

tco

mmercialized

byKU

KA.



Table2.

Summaryof

expe

rimen

tsforrobo

tsde

sign

edforscrew

insertion

Robo

tBriefde

scrip

tion

Evalua

tioncrite

riaRe

sults

Ope

ratedsegm

ents

Referenc

es

SpineA

ssist

Insertions

of32

guidewire

san

dfour

screwson

sixcada

vers.A

ccuracy

checke

dusingpo

st-ope

rativ

eCT

scan

s

Implan

ts’mea

suredde

viations

inall

plan

es(sag

ittal,c

oron

alan

daxial)

with

resp

ectto

thesurgical

plan

are<1.5mm

32of

36im

plan

tsco

rrectly

inserted

(88.89

%)

L1–-L5an

dS1

(18)

SpineA

ssist

Retrospe

ctivestud

yof

3912

screws

andgu

idewire

sinserted

onpa

tients

operated

in14

institu

tions,

June

2005

–June

2009

Surgeo

ns’judg

emen

tifpo

st-ope

rativ

eCT

scan

swereun

available.

Otherwise,

screwswereco

nsidered

correctly

inserted

if,at

most,brea

ches

of<2mm

wereob

served

With

outCT

:320

4of

3271

screws

correctly

inserted

(98%

).With

CT:6

35of

646screws

correctly

inserted

(98.3%

)

Una

vaila

ble

(19)

SpineA

ssist

Retrospe

ctivestud

yof

112pa

tients,

55op

erated

usingSp

ineA

ssistan

d57

with

aco

nven

tiona

lprotoco

l

Screw

completelyco

ntaine

dwith

inpe

dicleor,a

tmost,en

croa

ching

theco

rtical

bone

region

94.5

%usingSp

ineA

ssist.91

.5%

usingco

nven

tiona

lprotoco

lUnspe

cified

segm

ents

onthoracic,lum

baran

dsacral

region

s

(8)

Spineb

otv1

Target

follo

wingexpe

rimen

t,tracking

amov

ingtarget

which

emulated

the

patie

nt’sbrea

thing

Mea

suremen

tof

errorby

mea

nsof

anop

tical

tracking

system

Deviatio

nbo

unde

dby

!1.5mm

with

amaxim

umof

0.45

mm

Non

e(21)

Drillin

gof

amov

ingplastic

phan

tom,

usingtherobo

tas

tool

holder,a

ndthen

byco

mpletely

autono

mou

sdrilling

Errormea

suredas

distan

cebe

twee

nplan

nedan

dactual

perforation

points

usingapo

st-ope

rativ

eCT

scan

Obs

ervederrorin

the1–

2mm

rang

eforbo

thcases

Non

e

Spineb

otv2

Insertionof

screwson

14sp

inal

levels

oftw

ocada

vers

Screwsco

mpletelyco

ntaine

dwith

inbo

neswith

nobrea

ches.A

xial

and

laterala

nglesmea

suredusinga

post-ope

rativ

eCT

scan

26of

28screwsco

rrectly

inserted

(92.86

%)A

xial

deviationof

2.45

!2.56

" .Laterald

eviatio

nof

0.71

!0.21

"

T11–

T12,

L1–L5

(23)

VectorBo

tMachining

expe

rimen

tson

artifi

cial

bone

andabo

vine

spine,

observingthe

influe

nceof

vario

uspa

rameters,such

astool

choice,d

rillin

gsp

eed,

controlle

rga

ins,en

tryan

gles

andop

tical

tracke

rsamplingrate.N

oscrewswereinserted

Mea

suremen

tof

deviationerrors

betw

eenplan

nedan

dactual

entry

points.M

easuremen

tof

forceprofi

les

Millingbe

tter

than

drillingin

term

sof

forces

andaccu

racy.R

eactiveforces

always

<15

N.M

eande

viationformillingof

0.42

mm

with

amaxim

umof

1.7mm.H

ighsp

eeds

(3000

0rpm)a

ndsamplingrates(20Hz)

increa

seaccu

racy.H

ighprop

ortio

nala

ndintegral

gains

redu

cepo

seerrors

andsettlin

gtim

ewhile

increa

sing

therobo

t’sstiffne

ss

Non

e(25)

Neu

roglide

Screw

insertionon

thecervical

sections

ofsixcada

vers

Post-ope

rativ

eCT

scan

smea

surin

gthetran

slationa

land

angu

larerrors

ofinserted

screws

8of

10screwsco

rrectly

inserted

(80%

)with

mea

ntran

slationa

land

angu

lar

errors

of1.94

mm

and4.35

" *

C1–C2

(26)

*Rob

ot’spa

rametersweretune

ddu

ringexpe

rimen

ts,sono

tallinsertio

nswerepe

rformed

unde

rthesameco

ndition

s.Mea

nvalues

areno

trepresen

tativ

e.Briefd

escriptio

nsof

theexpe

rimen

tsaregiven,

aswella

stheused

crite

riafora

nalysis.Su

mmaryof

results

aregiven,

with

speciale

mph

asison

prop

ortio

nof

successfullyinserted

implan

ts(i.e.

screwsor

guidewire

s)an

dtheirob

served

deviations

A. Bertelsen et al.


MARS evolved into the SpineAssist, now commercializedby Mazor Robotics (Cesarea, Israel), which is still the onlyrobot available in the market, with FDA and CEclearances, specifically designed for spinal interventions,such as biopsies, transpedicular fixation, scoliotic backcorrection and vertebroplasty. To this date, SpineAssisthas been validated by 2500 procedures worldwide, onwhich over 15,000 implants have been placed withoutreported cases of nerve damage (15). Recently, MazorRobotics introduced the Renaissance, a new version ofSpineAssist, which, despite keeping its core technologies,had a complete overhaul in its software and user inter-face. It also added new features, such as the C-OnSite,which permits the acquisition of 3D images using anormal C-arm, by manually rotating it around the patient.The Renaissance is also expanding its clinical field beyondthe thoracic and lumbar spine: it has successfully beenused for brain biopsies and in the world’s first robot-assisted surgery on the cervical spine.2

MARS/SpineAssist was designed as an intelligent toolholder for interventions that required percutaneousinsertions of needles and screws. Its main innovation wasits reduced size and weight, which permitted its directattachment to the patient’s bony structure. This greatlysimplifies the registration on pre- and intra-operative images,as neither tracking nor immobilization are needed, as norelativemotion between the patient and the robot is possible.

The MARS prototype was a high-precision parallelmanipulator with six degrees of freedom (DoFs), whichhad a cylindrical shape (base 25 cm2#height 7 cm), aweight of 200 g, positioning errors of < 0.1mm and couldstand forces up to 10N (14). The team behind MARSadvocated the use of small robots, as they occupy less spacein the operating room. This, however, reduces their work-ing volume and makes them less able to withstand reactiveforces, which, in the case of drilling, can reach 15N.Although small robots are considered safer, as they are lessable to damage clinical staff in the case of a malfunction, itmust be noted that even small forces exerted on nerves orblood vessels can severely damage the patient.

SpineAssist, shown in Figure 1, is an improved version ofMARS, slightly bigger (base diameter 5 cm# height 8 cm,250g weight) and complemented with different mountingplatforms. For minimally invasive interventions, SpineAssistcan be used with the Hover-T, also shown in Figure 1, whichis a plastic railing anchored on two points of the patient’spelvis and one of the spinous process of an upper vertebra.For open procedures, the robot can be mounted directlyover the spine, using a clamp and bridge. In addition, twodifferent bed-mounted platforms are available for biopsies,cervical interventions and guided oblique lumbar interbodyfusion procedures, which are described later. It must benoted that SpineAssist suffers from its limited workingspace, so it may not be able to reach a required position dur-ing intervention, so additional extensions must be attachedto the mounting platforms to solve this problem (16–18).

The surgical workflow with SpineAssist consists of fivesteps: (a) planning of the optimal positions and dimensionsof implants, based on pre-operative CT scans; (b) attach-ment of the neededmounting platform to the patient’s bonyanatomy; (c) acquisition of two X-ray images, which areautomatically registered to the CT scan; (d) mounting ofthe robot on the platform, which latter aligns its arm withthe planned screw (or tool) trajectory; and (e) drillingthrough the guiding tube held by the robot’s arm, followedby the insertion of the guide wire and screw. Robot motion,drilling and insertion are repeated for all required implants.

After attainment of FDA clearance, SpineAssist has beenused by clinical teamsworldwide, mostly in Israel, Germanyand the USA, which have reported their experiences inmultiple peer-reviewed publications. In 2007, Togawaet al. (18) published the results of a cadaveric study usingSpineAssist for insertion of pedicle and translaminar facetscrews. For the first type of surgery, 32 guide wires and fourpedicle screws were inserted into six cadavers, evaluatingthe implants’ accuracy by post-operative CT scans. Theresults revealed that 32 of the 36 placements (88.89%)were within ! 1.5mm of the planned positions, with anoverall deviation of 0.87! 0.63mm. In 2010, Devito et al.

2http://ryortho.com/spine.php?news=1898_First-RobotGuided-Cer-vical-Spine-Surgery

Figure 1. (Top)SpineAssist robotwith an instrument guide. (Bottom)SpineAssist mounted over the Hover-T system. Reprinted withpermission fromMazor Robotics. Copyright©Mazor Robotics Inc.



http://ryortho.com/spine.php?news=1898_First-RobotGuided-Cervical-Spine-Surgery

http://ryortho.com/spine.php?news=1898_First-RobotGuided-Cervical-Spine-Surgery

(19) published a retrospective study about the use ofSpineAssist between June 2005 and June 2009 on 14 differ-ent hospitals worldwide, analysing a total of 842 patients.Intra-operative fluoroscopy was used to assess 3271 screwsand guide wires inserted in 635 patients, of which 3204(98%)were found to be correctly placed. In 139 cases,moredetailed quantitative analyses using post-operative CTscanswere made, which revealed that 635 of 646 implants(98.3%) were correctly inserted, with 577 (89.3%) beingcompletely contained within the pedicle and 58 (9%)showing breaches of < 2mm. No cases of permanent nervedamage were observed and 49% of the interventions weremade percutaneously, a considerably higher rate than thecommon 5% rate using non-robotic approaches cited bythe authors. In 2011, Kantelhardt et al. (8) published aretrospective study in which 55 patients underwent pediclescrew placement surgeries using SpineAssist and 57 wereoperated following a conventional protocol. The authorsfound that patients in the first group had a significantincrease in the proportion of screws placed with no breaches(94.5%vs 91.4%), reduced times of X-ray exposure per screw(34 s vs 77 s) and better recoveries after the interventions. Inaddition to the articles cited here, Mazor Robotics’ websitehosts an extensive list of publications about SpineAssist,which should be consulted by interested readers.3

SpineAssist has also enabled a new type of surgery, calledguided oblique lumbar interbody fusion (GO-LIF), whichconsists of the fixation of two vertebrae by insertion oftwo screws from the inferior vertebra to the superior one,crossing the intermediate inter-body disc space. In thisway, only two screws are needed for fixation instead ofthe typical four and the connecting rods. This type ofsurgery is inapplicable by conventional free-hand techni-ques, due to the high level of precision it requires. A recentpreclinical study on cadavers has demonstrated its feasibil-ity, with an error level of 1.3! 0.2mm with respect to thepre-operative plan (20).

SPINEBOT, SPINEBOT v 2 and CoRAThree different surgical robots have been presented byKorean researchers, all of them designed for transpedicularfixation. In 2005 a team from Hanyang Universitypresented the SPINEBOT, a robot capable of automaticdrilling, a feature missing from previously existing projectsfor this intervention (21). SPINEBOTused in-house planningsoftware and an optical tracking system based on sphericalreflective markers for localization of the surgical tools andpatient. In 2009, a team from the Pohang University ofScience and Technology (POSTECH) used SPINEBOT’splanning and tracking system with a different robot, calledthe cooperative robotic assistant (CoRA), a more robustprototype capable of automated screw insertion and hapticfeedback (22). In 2010, a cadaveric studywas reported usinga completely redesigned SPINEBOT (which will be named‘SPINEBOTv 2’), with fewer DoFs andwithout the automaticdrilling capabilities (23).

As mentioned in the previous paragraph, the originalSPINEBOT project comprised not only the surgical robotbut also planning software and an optical tracking system.The proposed software, named HexaView, allowed thesurgeons to plan the screw insertions using six differentviews of a CT or magnetic resonance (MR) scan of thepatient. The optical tracking system was commercializedby NDI (Waterloo, Ontario, Canada) and offered feedbackat 30Hz for redundant position control of the robot, inaddition to its embedded encoders. The robot, as shownon Figure 2, consisted of a Cartesian positioner, a gimbaland a tool holder, which provided three, two and two DoFs,respectively, giving seven DoFs in total. The robot was ableto do the gross and fine positioning of the surgical toolsand keep them in place while the surgeon carried outthe drilling, although it was capable of doing this taskautonomously if desired. SPINEBOT also included amotion-correction system, a remarkable featuremissing fromearlier projects, which was based on the optical feedbackand could correct the patient’s motion produced by breath-ing, which had an amplitude of 3mm in the antero-posteriordirection according to the authors (21).

Chung et al. (21) reported multiple experiments usingSPINEBOT. In one, they made the robot follow a movingtarget, which performed a sinusoidal motion with an ampli-tude of 2mm and a period of 5 s, emulating a breathingpatient. SPINEBOT was able to follow the target with anerror bounded by ! 0.15mm and a maximum of 0.45mm,values close to the 0.35mm error introduced by the opticaltracking system. In a second experiment, holes were drilledin a moving plastic phantom, first by a person who used theSPINEBOTas a tool holder and then by the robot working infully autonomous mode. In both cases, observed deviationswere in the 1–2mm range, although the robot’s accuracyseemed slightly superior.

Lee et al. (22) used the SPINEBOT’s planning and track-ing systems with the improved robot CoRA, a sophisticateddevice capable of automated drilling and screw insertion, afeature which, according to the authors, was implementedfor the first time. The robot, as shown in Figure 3, was builtwith a more robust frame, which permitted it to withstandlarger reaction forces but hindered the surgeon’s access tothe patient. In addition, CoRA offered cooperative control,a tele-operated drilling system with realistic haptic feed-back and a small and lightweight end-effector. Lee et al.published some proof-of-concept experiments in theirarticle, but did not provide a quantitative analysis of CoRA’sperformance. As the planning and tracking systems wereidentical to SPINEBOT’s, the authors expected CoRA tohave similar levels of error in screw placement (1–2mm)as the main source of inaccuracies, -that is, the opticalsystem- remained unchanged. At the time of writing thispaper, no further experiments on phantoms or cadaversusing CoRA have been published.

The SPINEBOT v 2, presented in 2010, was completelydifferent from the first SPINEBOT, despite keeping thesame name. As can be seen in Figure 4, the new robothad only five DoFs – one prismatic and four rotationaljoints – and, more importantly, it lacked the automated

3http://www.mazorrobotics.com/int/physicians-int/peer-review-library-int.html

A. Bertelsen et al.


http://www.mazorrobotics.com/int/physicians-int/peer-review-library-int.html

http://www.mazorrobotics.com/int/physicians-int/peer-review-library-int.html

drilling capabilities, replacing the original end-effector bya simpler tool holder. Its planning software also wentthrough a complete redesign and the tracking systemwas replaced by biplanar continuous fluoroscopy. Insteadof relying on optical tracking, SPINEBOT v 2 detected thepatient’s and tool positions by processing the fluoroscopicimages – updated at 20Hz – and by means of custom2D–3D registration algorithms (23).

The authors performed laboratory experiments to esti-mate the overall positioning error of SPINEBOT v 2, whichwas found to be 1.38! 0.21mm. On the cadaveric tests, 28screws were inserted in 14 different vertebrae of two

cadavers and post-operative CT scans were made to assessthe screws’ insertions and measure their angular deviations.Of the 28 total screws, 26 were correctly positioned (successrate of 92.86%), with no observed perforations into thespinal canal. Average angular errors were 2.45! 2.56" and0.71!1.21" in the axial and lateral planes, respectively (23).

VectorBot/KinemedicThe Deutsches Zentrum für Luft- und Raumfahrt (GermanAerospace Centre, DLR) has developed a series of light-weight robots designed for multiple surgical scenarios,giving a desirable degree of versatility that could compensate

Figure 3. Schematic of the CoRA proposed by Lee et al. (22), withits dimensions and DoFs. Reprinted with permission fromHapticsand Virtual Reality Laboratory, Pohang University of Science andTechnology (POSTECH). Copyright © Haptics and Virtual RealityLaboratory, Pohang University of Science and Technology(POSTECH)

Figure 4. SPINEBOT v 2, as described (23). The treadmill atbottom right corresponds to the robot’s prismatic joint.Reprinted with permission from Centre for Intelligent SurgerySystems, Hanyang University. Copyright © Centre for IntelligentSurgery Systems, Hanyang University

Figure 2. The first version of SPINEBOT described (21): (left) the whole system, showing the three DoFs of the Cartesian mechanism;(top right) close-up of the gimbal which provides two DoFs and holds the tool guide; (bottom right) close-up of the linear tool guide,with its additional two DoFs. Reprinted with permission from Centre for Intelligent Surgery Systems, Hanyang University. Copyright© Centre for Intelligent Surgery Systems, Hanyang University



the increasing cost and complexity of medical roboticsystems. DLR’s research has also covered spinal surgery,specifically transpedicular fixation, as described in publica-tions by Ortmaier et al. (24,25). This project was giventhe name ‘VectorBot’ by BrainLab (Feldkirchen, Germany),who sponsored it with a $5 million investment but, unfortu-nately, cancelled it before its introduction to themarket (13).

The VectorBot consisted of the DLR’s Kinemedic robotcoupled with the VectorVision optical tracking systemdeveloped by BrainLab, although early prototypes weredeveloped using the preceding Light-weight Robot II(LWRII). Both robots are shown in Figure 5. TheVectorBot required no X-ray images, as all the trackingwas made using markers attached to the patient’svertebrae and points collected by the optical system. Thus,radiation was reduced to a minimum but at a cost ofincreased invasiveness, due to the large incisions requiredto expose the spine. In line with other research projects,the authors preferred a robot that worked as an assistant,providing help to the surgeon rather than executing theintervention autonomously. This assistance came inthe form of virtual fixtures, i.e. physical limits imposed bythe robot, which prevented the surgeon from deviatingtoo much from the planned trajectories. The robot was notcapable of automatic drilling or screwing, although it keptthe surgical instrument in a safe and stable location whilethe surgeon remained in control of these tasks.

In 2006, Ortmaier et al. (25) published the results of aseries of evaluation experiments with their proposedsystem. In them, the authors carried out two differentmachining tasks, drilling and milling, on a block of artificialbone and a bovine spine, measuring hole diameters, poseerrors and reactive forces for different machining tools,entrance angles and values of control parameters. Summa-rizing their results, they concluded that milling wassuperior to drilling in terms of deviation errors and reactiveforces, due to the larger slippage of the drill tip observedduring drilling, which bent the instrument and increasedfriction inside its guide. Mean deviation error for millingin the plane perpendicular to the instrument axis was of0.42mm, with the maximum reaching up to 1.7mm.

Maximum forces reached up to 15N, well below the limitof 30N which could be handled by the robot. In terms ofcontrol parameters, the authors concluded that theoptimum was reached with high proportional and integralgains, which led to higher robot stiffness, lower pose errors,reduced settling time and decreased overshoot. The authorsidentified the accuracy and latency of the optical trackingsystem as critical factors. However, they acknowledgedthat additional sources of error in the system were nottaken in account in their study, such as pre-operative imageresolution, segmentation accuracy and intra-operativeregistration errors.

NeuroglideThe majority of robots for screw insertion were designed tooperate in the lumbar section. This has technical advan-tages – lumbar pedicles are larger than thoracic and cervicalones, so precision requirements are less demanding – butalso clinical relevance, as fusions in the lumbar area aremore common than those carried out in other spinalregions. This has reduced interest in interventions at thecervical level, a problem addressed by Kostrzewski et al.(26), who proposed theNeuroglide4 robot for cervical inter-body fusion in 2012. The proposed system was designedspecifically for atlanto–axial fusion, i.e. the fusion of theupper two vertebrae, C1 and C2, by means of screwsinserted through both of them.

The Neuroglide consisted of a high-precision, parallel,four-DoF mechanism that held a drill guide, as shown inFigure 6. The robot’s reduced size limited its workspace,but gross positioning was carried out by means of a passiveserial arm on which the robot was mounted, also shown inFigure 6. Navigation was implemented using an infra-redoptical tracker and active markers attached to the robotand vertebrae, previously exposed by an incision andregistered to the pre-operative space by probing points onthe bony surface. The authors also developed a custom

Figure 5. The DLR’s surgical systems: (left) the prototype using the LWR II, coupled with the navigation system and reflectivemarkers; (right) Kinemedic robot, successor of the LWRII. Left figure reproduced from (25). Reprinted with permission from JohnWiley and Sons. Copyright © 2006 John Wiley and Sons Ltd; right figure, Reprinted with permission from Institute of Robotics andMechatronics. Copyright © Institute of Robotics and Mechatronics, DLR

4Kostrzewski et al.’s publication does not cite the robot’s name, whichwas taken from the project’s website (http://lsro.epfl.ch/vrai/pro-jects/Neuroglide)

A. Bertelsen et al.


http://lsro.epfl.ch/vrai/projects/Neuroglide

http://lsro.epfl.ch/vrai/projects/Neuroglide

joystick for robot control and software for navigation andsurgical planning, which was used to determine the screwtrajectories based on pre-operative CT scans.

The Neuroglide was evaluated in a feasibility experimentwith six cadavers, in which an experienced neurosurgeoninserted a total of 10 screws, fusing the C1 and C2vertebrae, and then evaluated the insertions with post-operative CT scans. The mean translational error reportedby the authors was 1.94mm and the mean rotational errorwas 4.35", although two screws were dropped from thestatistical sample, due to their abnormally large errorsproduced by drill slippage. It must be noted that the authorsimproved multiple aspects of their system while the experi-ments were under way, so results are not comparabledirectly, as they were not obtained under the same condi-tions. Furthermore, the sample size was not large enoughto draw meaningful conclusions. However, the authorsreported a remarkable result after all their improvements

were in place (0.41mm and 2.56" for the last screw) andplanned further cadaver testing. In addition, Kostrezwskiet al. measured the average time needed to use their systemand claimed that a conventional image-guided procedurewas only 3min shorter, a negligible difference for anintervention lasting several hours.

RIMEBoschetti et al. (27) in 2005 proposed the robot in medicalenvironment (RIME) project, a robotic system designedfor drilling in transpedicular fixation surgeries. Theproject’s main contributions were development of a fullyteleoperated system, which permitted the surgeon tooperate on a patient who could be kilometres away, andhaptic feedback, provided to the surgeon using the customPiRoGa5 device. Experiments reported in 2007 by Rosatiet al. (28) demonstrated the feasibility of haptic feedbacktransmission and control of a six-DoF industrial robotbetween two cities separated by 35km, although theauthors still needed to integrate the optical tracking deviceproposed by Boschetti et al. into the whole system. Nopublications about experiments with cadavers or animalswere found at the time of writing of this study.

RSSSJin et al. (29) have recently proposed a new surgical robotfor pedicle screw insertion, named the robot spinal surgicalsystem (RSSS), based on a five-DoF SCARA robot equippedwith an infrared tracking device. RSSS’s mechanical designensures that the robot should not collapse under its ownweight in the case of a power failure, ensuring the patient’ssafety. RSSS offers haptic feedback, virtual fixtures, a screw-implanting mechanism and a control strategy for auto-mated drilling, which is able to identify the force profilesfor each drilling stage and automatically stop beforebreaching the vertebra (30). Currently, this project is atan experimental stage and only experiments for the tuningof control parameters have been reported.

Robot for laminectomy

In 2010,Wang et al. (31) proposed a robot for laminectomy,i.e. removal of posterior bony sections of vertebrae to allevi-ate nerve compression produced by diseases such asstenosis. This type of procedure requires milling of bone inthe vicinity of the spinal cord; thus, a high level of precisionis required, as damage to the latter must be prevented at allcosts. The authors proposed a robot with two translationalDoFs capable of automatic machining of the lamina andable to stop just before penetration into the spinal canal,leaving a thin layer of bone which could be later removedby the surgeon. The robot was equipped with force sensorsand custom algorithms able to identify the bone layer beingmachined, according to the measured force profiles. Theauthors reported experimental results on 10 bovine spinesamples, in which they measured the thickness of the bonelayer left by the robot, which had an average value of1.1mm. No breaching into the spinal canal was observedand the robot’s recorded working times were 10–14min,

Figure 6. Neuroglide robot for cervical surgery: (top) aspect of thecomplete system with the robot (R), passive mounting structure(PS), optical tracker (T) andMayfield clamp (M) for skull attachment;(bottom) illustrationof the robotwith the drill holder coloured in red.Robot’s DoFs are along the y and z axes and around angles a and b.Figures reproduced from (26). Reprinted with permission from JohnWiley and Sons. Copyright © 2011 John Wiley and Sons Ltd



similar to times taken by surgeons doing the same task.Further work was planned by the authors to build amore stable mounting platform for the robot, as well asadditional experiments.

Robots for needle-based interventions

A considerable number of recent robotics projects (5 of 18)have addressed needle-based interventions, such asbiopsies and vertebroplasties. As it will be seen, many ofthese were not specifically designed for use on the spine,as needle-based interventions can be executed in manyother anatomical regions. A summary of the most relevantexperiments using these robots is given in Table 3 anddetailed descriptions are given in the sections below.

AcuBotOne year before publication of theMARS robot, Cleary et al.(32) presented a plan for development of a minimallyinvasive system for spinal surgery. In this article, theauthors identified multiple technical problems found inthe implementation of a system of this kind: unavailabilityof intra-operative axial images, difficult fusion of CT andMR data, lack of visualization of oblique trajectories,unavailability of spinal tracking systems, slow and difficultinstrument insertion and lack of appropriate software.The authors, to solve the aforementioned problems, advo-cated the use of intra-operative CT, 3D visualization, opticaltracking systems, robotic tool holders and development ofspecialized software. This technical plan led to thedevelopment in 2003 of the AcuBot, a robot designed forpercutaneous needle insertion guided by fluoroscopy, usingone or two planes, or intra-operative CT scans (1). AcuBotreceived clearance from the FDA and clinical trials with20 patients who underwent spinal nerve blockade werepublished in 2005 (5).

Acubot, shown in Figure 7, had a total of six DoFs,decoupled in positioning, rotation and instrument inser-tion. A three-DoF Cartesian manipulator, mounted on abridge attached to the scanner’s table, supported a remotecentre of motion (RCM) mechanism with two DoFs, withthe instrument mounted on its end-effector, which had anadditional translational DoF along the instrument axis.Normally, the Cartesian manipulator brought the needleclose to the insertion point, the RCM corrected its orienta-tion and, after approval of the surgeon, the instrumentwas inserted to reach the target point inside the patient’sbody. In addition, a seven-DoF passive mechanismwas mounted between the Cartesian manipulator and the

Table 3. Brief description of reported experiments for robots designed for needle insertion tasks

Robot Brief description Evaluation criteria Results Reference

Acubot Randomized clinical trial with20 patients, of which 10 wereoperated with the robot and10 using a conventional manualtechnique

Measurement of needledeviation from the plannedtarget point usingbiplanar fluoroscopy

Mean deviation with robot,1.105mm; mean deviation withmanual technique, 1.238mm

(5)

Innomotion Testing of robot-guidedpercutaneous needle insertionson four pigsput under general anaesthesia

Target deviation measuredusing MRI

Axial deviation in the !1mmrange (min, 0.5mm; max, 3mm).Transverse angular deviation inthe !1" range(min, 0.5"; max, 3")

(2)

Innomotion 25 MR-guided punctures onphantoms placed withinwater-filled containers

Deviations from plannedtarget points measured byhand, using rulers

Observed deviation of2.2!0.7mm

(35)

DLR LWR III In vitro mechanical trials usinga precisely constructed phantom

Deviation measured frompost-operative 3D radiographies

1.2!0.4mm (max, 1.98mm) (3)

(Onogi et al.) In vitro test of 50 punctureson the pedicles of fivespinal phantoms

Deviation measured withpost-operative CT scans

Translational deviation:1.46!0.80mm; angulardeviation, 1.49!0.64"

(36)

SpineNav In vitro mechanical trials,measuring errors in 30positions within therobot’s workspace

Deviation measured usinga 3D digitizing arm

0.89mm (max, 1.14mm) (38)

Figure 7. AcuBot being used in a CT-guided needle insertion: thethree DoF Cartesian manipulator is placed over the black bridgeattached to the scanner’s table; the RCM and the linear end-effector are placed at the end of the passive articulated arm,which comes out from the right side of the Cartesian mechanism.Reprinted with permission from URobotics Lab, Johns HopkinsUniversity. Copyright© 2012URobotics Laboratory, Johns HopkinsUniversity

A. Bertelsen et al.


RCM mechanism, which was manually adjusted by thesurgeon before the procedure, bringing the instrumentclose to the entry points. The AcuBot also included adisplay and a joystick, used by the surgeon to control therobot remotely.

A clinical trial was reported by Cleary et al., in which theAcuBot was used to perform biplanar fluoroscopy-guidednerve and facet blocks (5). This type of procedure consistsof the localization of the source of back pain by insertionof 22-gauge needles in precise locations of the spine,followed by injection of local anaesthetics. In the reportedtrial, a randomized studywas performed at the GeorgetownUniversity Medical Centre with 20 patients, 10 of whichunderwent the conventional procedure and the remaining10 were operated using the AcuBot. Needle insertionaccuracy and pain relief were measured for both groups ofpatients and the results were similar for both measures.Mean deviations for the robot and manual methods were1.105 and 1.238mm, respectively. Pain scores, measuredin the range 0–10, were reduced from 6.3 to 1.8 using therobot and from 6.0 to 0.9 using the manual method.Although the results give the impression that the AcuBotis as accurate and effective as the manual method, theauthors acknowledged that the statistical sample was toosmall to draw significant conclusions. Currently, researchwith the AcuBot is focused on development of a rotatingneedle holder for improved lesion targeting in soft organs,such as the liver and lung, giving less attention to spinalprocedures (33).

InnomotionMR offers interesting advantages for robotic and image-guided surgery, as it offers superior soft tissue contrastand does not irradiate the patient. However, the strongmagnetic fields present in all MR systems greatly compli-cate the robotic design, as compatible materials, sensorsand actuators must be used. In addition, an MR-compatiblerobot should have a reduced size to fit into the small spaceof the magnet’s bore, largely occupied by the patient. Allof these factors make development of MR-compatiblerobots a formidable challenge. A remarkable example, usedin spinal procedures, is the work by Hempel et al. (34), whoin 2003 presented the manipulator for interventionalradiology (MIRA), which evolved into the Innomotionsystem (2). The latter obtained CE clearance and was intro-duced to the market by Innomedic GmbH (Herxheim,Germany), which was acquired by Synthes (Solothurn,Switzerland) in 2008. The Innomotion’s commercializationwas stopped in early 2010 and is expected to be restarted in2012 by the IBSmm Company (Brno, Czech Republic),which is now working on the robot’s improvement.

Innomotion, shown in Figure 8, was designed as a tele-manipulator for MR-guided insertion of cannulae andprobes for biopsy, drainage, drug delivery and energetictumour destruction. Although direct interventions in thecentral nervous systemwere left out, due to the demandingregulations and long approval process, Innomotion can stillbe used for interventions in the spine’s periphery. Therobot’s kinematics consists of an arm driven in five DoFs,

attached to an orbiting ringmounted on the scanner’s table,and is equippedwith linear pneumatic actuators and opticallimit switches, rotational and linear encoders. Innomotion’sinstrument holder was designed as a RCM with two DoFsand was equipped with gadolinium-filled spheres, whichcould be easily segmented from intra-operative MR imagesto detect its position and orientation.

Melzer et al. (2) published the results of animal testsin which the robot’s deviation in the axial plane wasestimated to be within the !1mm range (minimum,0.5mm; maximum, 3mm) and its angular deviation to be1" (minimum, 0.5"; maximum, 3"). These results werecompliant with the CE standard, although they were notsufficient for interventions in the central nervous system(2) In 2010, Moche et al. (35) published a study withaccuracy measurements of Innomotion, using phantomsas well as clinical trials. In the phantom study, 25 needleinsertions were performed, with observed deviationbetween the target and observed points of 2.2! 0.7mm,measured by hand using rulers. The reported clinicaltrials consisted of diagnostic biopsies which requiredplanning and execution times of 25 and 44min,respectively. Of the six reported interventions, two werecarried out successfully around the spine: one was a bonebiopsy in the iliac crest and one was an abscess aspirationin the L5–S1 region. No complications were observed inall six cases.

DLR’s LWR IIIA following version of DLR’s LWR, the LWRIII, is now com-mercialized by KUKA and is increasingly being adopted forsurgical robotics projects, such as the one designed for spinebiopsies and vertebroplasties presented by Tovar-Arriagaet al. (3). This project consisted of the aforementionedrobot guided by intra-operative 3D radiographies, acquiredby a rotational C-arm, and an optical infra-red trackingsystem. The authors reported two experiments. In the first,they measured the errors of calibration between the tooltippositions measured by the optical system and the robot’s

Figure 8. Innomotion robot, used for CT- and MR-guidedneedle-based procedures. Reprinted with permission fromInnomedic GmbH. Copyright © Innomedic GmbH



controller, which had a mean of 0.23mm, a deviation of0.1mm and a maximum value of 0.47mm. In the secondexperiment, the authors positioned the tooltip in variouslocations over a precisely manufactured phantom andmeasured the deviations with 3D radiographies; theauthors estimated the error to be in the 1.2! 0.4mm range,with minimum and maximum values of 0.3 and 1.98mm,respectively. Overall, the reported accuracies were accept-able for the demands of surgery, although the authors citedthe optical system’s accuracy and low sampling rate(20Hz) as limiting factors that should be improved infuture versions.

University of Tokyo’s robot for vertebroplastyA group of Japanese researchers from the cities of Tokyoand Osaka presented a robotic system for vertebroplasty,based on a robot with a compact end-effector, which couldbe inserted in the space between the C-arm and the patient.Surgical planning was carried out on pre-operative CTscans, whilst intra-operative guidance relied on fluoros-copy. For this purpose, the needle holder was built withplastic material to make it partially radiolucent. In addition,this mechanical device could be automatically detachedfrom the robot by a safety mechanism, triggered whenexcessive forces were applied to the needle (4). In 2009,Onogi et al. (36) reported an in vitro experiment in whichthe robot was used for 50 punctures in the pedicles of fivepolyurethane phantoms of lumbar spines. Deviation,measured from post-operative CT scans, was estimated tobe 1.46! 0.80mm and 1.49! 0.64" (36).

SpineNavIn 2008, Ju et al. (37) presented the SpineNav, a robot forpercutaneous vertebroplasty which could insert needlesautonomously or using a tele-operatedmechanismwith fiveDoFs. This robot was designed to be used inside a CTscanner and its mounting platform has a metal mask whichcan be easily segmented from the intra-operative images toestimate the robot’s base position and orientation withrespect to the patient. Accuracy tests carried out by theauthors estimate SpineNav’s mean positioning error as0.89mm, with a maximum of 1.14mm (38). To date, noreports about experiments with cadavers or clinical trialsusing SpineNav are available.

Robots for endoscopic interventions

MINOSC – sub-arachnoid space explorationThe microneuro-endoscopy of spinal cord (MINOSC)European project led to the development of a roboticsystem for interventions of the spinal cord from withinthe sub-arachnoid space. This is a challenging task, as thissection of the spine is only a few millimetres wide and issurrounded by delicate structures which can easilybecome damaged. Ascari et al. (9) published results ofthe design of a robot-assisted endoscope, which providesthe surgeon with direct vision of the surrounding struc-tures – spinal cord, blood vessels and nerve roots – andpermits operations such as localized electro-stimulation.

The system uses image-processing techniques to analyseits surroundings and give feedback to its control unit,which can steer the endoscope tip to avoid obstacleswhich may not even be present in the endoscope’s fieldof view. Steering is implemented by a two-DoF cable-driven mechanism and three lateral hydraulic jets thatstabilize the endoscope’s tip.

In 2010, Ascari et al. (9) reported a series of in vitro,ex vivo and in vivo experiments, which validated all theprototype’s subsystems, excluding navigation, which wastested up to in vitro experiments. In addition, localizedelectrostimulation of nerve roots was successfully accom-plished in an additional in vivo test. According to theauthors, the prototype is still far from reaching clinicaluse, but the major implementation problems were alreadysolved in its current stage.

da VinciThe da Vinci surgical robot (Intuitive Surgical, Sunnyvale,CA, USA), mostly used in urological and gynaecologicalsurgeries, has also been tested for endoscopic spinal inter-ventions, although its applications are limited. In fact, theda Vinci’s end-effectors are not well suited for bone drilling,due to the limited range of force they offer, as they weredesigned primarily for the manipulation of soft tissue(39). However, there are reports of successful experimentsusing it for spinal interventions, although all are at an earlyexperimental stage.

Yang et al. (7) published a review of experimental uses ofthe da Vinci on spinal procedures, along with a report offive successful cases of paravertebral tumour resections.Lee et al. (40) published a study on two cadavers, whichdemonstrated the feasibility of using the da Vinci for trans-oral decompression of the cranio–cervical junction, andPonnusamy et al. (41) reported successful laminotomy,laminectomy, disc incision and dural-suturing procedureson a pig, using a posterior approach. The lack of appropriatetools for the da Vinci is repeatedly cited as a problem,although Ponnusamy et al. reported the use of a prototypeburr, rongeur and laser instrument, the last one used forrapid coagulation.

Kim et al. (42) reported an experiment on anteriorlumbar interbody fusion (ALIF) using the da Vinci on apig, inserting a metal cage in the inter-disc space. AlthoughALIF was proposed several years ago, post-operativecomplications have prevented its widespread use. The workof Kim et al. (42) expects to increase this surgery’s safetyby incorporating robotic assistance, although it is still atan early stage.

Radiosurgery robotic systems

Current radiosurgical systems employ heavy-duty robotsto move a linear accelerator (LINAC) around the patient,firing high-energy beams according to a predefined plan,ablating internal tumours and minimizing damage tosurrounding healthy tissue. Although radiosurgical systemsdo not come into contact – or even near – the patient, we

A. Bertelsen et al.


considered that they should be included in this review, asthey are currently used for the treatment of spinal lesionsand they fit the criteria set out in Methods, as they auton-omously perform a clearly defined surgical task with directimpact on the patient’s body.

Radiosurgery was conceived as a treatment for deep-seated intracranial tumours, for which conventionalsurgery is considered too dangerous or infeasible. Thefirst commercially available radiosurgical system was theGammaKnife (Elekta AB, Stockholm, Sweden), whichwas introduced to the market in 1968, and since then,radiosurgery has gained worldwide acceptance and isnow considered within standard oncological practice.Nowadays, the market is dominated by the CyberKnife(Accuray Inc., Sunnyvale, CA, USA) and the Novalis(BrainLAB, Heimstetten, Germany), which permit inter-ventions guided by intra-operative imaging without theneed of stereotactical frames. Using this, a pair of X-raydevices acquires images of the patient at regular intervals,which are processed to monitor the patient’s position andadjust the LINAC accordingly, minimizing deviations fromthe surgical plan.

Nowadays, radiosurgery is used for the treatment oflesions found in many extracranial regions, includingthe spine. A recent review by Romanelli and Adler (6)cites multiple clinical studies of spinal radiosurgery, stat-ing that this technique is well suited for treatment ofneoplastic lesions and intramedullary arteriovenousmalformations. In addition, spinal radiosurgery offers aneffective and well-tolerated option, as suggested by astudy made in Pittsburgh, which followed 393 patientsand observed high rates of long-term pain control (86%)and long-term tumour control (88%), with no cases ofneurological damage induced by radiation (43). Radiosur-gery was not considered practical before the introductionof image guidance, as the first reported cases used astereotactic frame which had to be attached to the spinousprocesses of the vertebrae, previously exposed by multipleincisions. This was naturally too cumbersome and notusable in treatments that required multiple radiation dosesdistributed along multiple sessions (44). The introductionof frameless image-guidance permitted more practical usesin spinal surgery, although the first reported studies stillrelied on fiducial markers inserted into the vertebrae adja-cent to the lesion. Further development of image-guidancetechnology permitted interventions without the need ofany type of marker and without noticeable reductions inaccuracy: studies estimate that fiducial-based spinal radio-surgery is accurate to within a mean distance of 0.7mm(45) and image-based is accurate to a mean distance of0.5–0.6mm (46,47).

Accuray and BrainLAB host extensive lists of publica-tions related to the CyberKnife5 and Novalis6 systems ontheir respective websites. Interested readers shouldconsult them for additional information.

Discussion

Robot design and safety

In the early days of surgical robotics, researchers adaptedindustrial robots for use in the operating room but, in thelast decade, there is a clear tendency in favour of specificallydesigned ones. Spinal surgery has been no exception. In theearly years, 1992–2002, researchers such as Sautot et al.(10) and Santos-Munné et al. (11) adapted industrial robotsfor surgical use. In the last decade, we have seen the appear-ance of robots specialized for surgical applications. Remark-able examples are SpineAssist (14), AcuBot (1), Innomotion(2), the SPINEBOTseries (21,23), CoRA (22) and the DLR’sLWR series (3,24). More recent studies continue in thesame line, proposing new models rather than adaptingindustrial robots. Examples are SpineNav (37), RSSS (29)and the studies by Onogi et al. (4). This shift has beencaused by safety requirements: industrial robots aredesigned to perform tasks – usually involving high torqueor speed – in the absence of humans, whereas surgicalrobots must constantly interact with the surgeon, clinicalstaff and the patient (who is absolutely unable to react incase of emergency). Besides, it must be possible to makethem sterile, using draping to reduce the risk of infection.It is noteworthy that, as medical robots are relatively new,no international standards yet exist for them, althoughthe International Standardization Organisation (ISO) andthe International Electrotechnical Commission (IEC) arecurrently working on their development.

There is also a tendency to increase safety by giving lessautonomy to robots in the operating room. In fact, surgeonsseem to prefer to be in control of all the intervention’s tasks,restricting the use of robots for assistance. This approachis desirable, as it combines the strengths of robots (stability,precision and immunity to fatigue) and humans (betteranalysis, judgement and response in unexpected situa-tions), who work cooperatively and increase the surgery’ssafety. A remarkable example is the SPINEBOT series: itsfirst prototype had automated motion and drilling (21),but these features were not present in the second model.In fact, SPINEBOT v 2’s main capability was keeping theinstrument in a stable position, leaving gross positioningand drilling in hands of the surgeon (23).

Accuracy

A summary of accuracy experiments for robots designed forscrew insertion is given in Table 2, which shows that manyare capable of inserting screws with > 85% possibility ofsuccess and a deviation of 1–2mm. Of particular interestis the retrospective clinical study by Devito et al. (19), whichanalysed 646 insertions performed using SpineAssist andconcluded that 635 (98.3%) were inserted with errors< 2mm. Table 3 also shows a summary of experiments,but only for robots designed for needle-based procedures.The table’s data shows that these robots are also capableof precise insertions with errors bounded by the same

5http://www.accuray.com/healthcare-professionals/clinical-publica-tions/cyberknife-publications#spine6http://www.novalis-radiosurgery.com/wp-content/uploads/2009/07/NTx+NV_Biblio_DEC0709_SPINE.pdf



http://www.novalis-radiosurgery.com/wp-content/uploads/2009/07/NTx+NV_Biblio_DEC0709_SPINE.pdf

http://www.novalis-radiosurgery.com/wp-content/uploads/2009/07/NTx+NV_Biblio_DEC0709_SPINE.pdf

values. It can be concluded that current robotic technologyis capable of accurate instrument placement in the 1–2mmrange, but not yet able to reach sub-millimetre accuracy inrealistic conditions (i.e. in actual interventions or in vivoexperiments). In the current situation, we could say thatrobots are faced with a ‘1mm barrier’, which they havenot yet been able to overcome. Going below this limit hasclinical relevance, as pointed out by Rampersaud et al.(48), who estimated that 1mm and 5" were the maximumtranslational and rotational errors that could be admitted inscrew insertion in the mid-cervical spine, the mid-thoracicspine and the thoracolumbar junction. Among thefactors that influence this, we can cite imaging systemresolution, registration inaccuracies and the motion ofvertebrae. Robots for screw insertion are also affected bydrill slippage, vibrations and reactive forces, whereas themain sources of error for needle-based ones are needledeflection and tissue deformation.

Registration and tracking technologies

In terms of tracking and registration technology, projectsdesigned for screw insertion have preferred optical track-ing, with the remarkable exceptions of SpineAssist andSPINEBOT v 2. Robots for biopsies and other needle-basedprocedures prefer less invasive tracking technologies, suchas fluoroscopy and intra-operative CT, which preventunnecessary incisions. Today, all available technologies fortracking and registration have inconveniences. On the onehand, optical tracking offers sub-millimetre precision ata reasonable cost and can reduce the required number ofX-rays, as periodic imaging to locate the surgical toolsbecomes less necessary. However, optical tracking suffersfrommarker occlusions and low sampling rates – somethingwhich has been repeatedly cited as a problem (3,25) – andmounting of markers requires rigid attachment to thepatient’s bony structures, which translates into moreincisions and higher invasiveness. On the other hand,fluoroscopy and intra-operative CTare less invasive, but theiruse exposes the patient, surgeon and clinical staff to higherdoses of radiation, which is undesirable. Intra-operativeMR is not well suited for spinal surgery, as bone is not visibleon MR images, as it produces no signal. Besides, it is stilluncommon in most hospitals and its strong magnetic fieldmandates the replacement of all surgical tools and implantsby their MR-compatible versions.

Patient registration has the additional problem of therelative motion of the vertebrae. If a single vertebra istracked, it is unrealistic to assume that the adjacent oneswill move in the same manner or, in other words, toconsider the spine as a rigid body. One option to solve thisproblem is reduction of the spine’s range of motion bymeans of additional hardware, such as the SpineAssist’sHover-T. Another interesting approach is the one used bySPINEBOT v 2, which uses individual registration ofvertebrae from biplanar fluoroscopic images updated inreal time, which, obviously, has the inconvenience ofhigher radiation (23). The spine’s relative motion cited

here and the tracking systems’ trade-offs make the choiceof a registration strategy a problem for which no obvioussolution yet exists.

Project development

Nowadays, Mazor Robotics’ SpineAssist is the onlycommercially available robot –with FDA and CE clearances– specifically designed for spinal surgery, particularly trans-pedicular fixation. There are seven other projects for similarapplications but, as far as we know, none of them have yetobtained any of the aforementioned clearances. The diffi-cult and expensive certification procedures and the highcost of robotics projects per se are big hindrances for compa-nies wishing to enter the medical robotics market. All thisforms a scenario in which Mazor Robotics could remainwithout direct competitors for some time.

However, other commercial robots are now used forspinal surgery, although they, differing from SpineAssist,were not specifically designed for this application. In fact,CyberKnife and Novalis are used for spine-related proce-dures and the da Vinci has also been used for them,although the reported cases are still at experimentalstages and its adoption in a clinical environment doesnot seem to happen soon.

Economic analysis

To the best of our knowledge, there is no publication avail-able which analyses the costs and benefits, from the clinicalpoint of view, of robotics in spinal surgery. Evenmore, thereare no studies today regarding the cost-efectiveness ofimage-guided spinal surgeries, according to Tjardes et al.(49). It is necessary for hospitals to ensure that the benefitsbrought by robotic surgery outweigh the costs, which arevery high. As an example, installation of a Mazor Robotics’Renaissance system has a cost of $789K (including therobot, workstation, instrument tray and 1 year of technicalsupport), while each intervention requires disposablematerials, which cost $1.2K, and implants valued between$6K and $8K. In addition, technical support from Mazormust be renewed annually, signing contracts of 10% ofthe installation price per year (15).

Conclusions

This article has reviewed the state of the art in surgicalrobotics for spinal interventions. Several prototypes andcommercially available systems have been analysed,showing that this particular field is still at an early stageof development. Up to this date, only one robotspecifically designed for spinal surgeries is available inthe market –SpineAssist/Renaissance – while the othersare research prototypes or commercial robots originallydesigned for other uses.

A. Bertelsen et al.


Among spine-specific applications, the most studiedone has been screw insertion, for which current technol-ogy offers increased levels of safety, considerably reducingthe number of misplaced screws. The accuracy of robotsfor this application, and also for needle insertion, permitsinstrument placements with deviations of 1–2mm.

Robots not only make existing surgeries safer: they alsoenable surgeons to do interventions which, without theirassistance, would be absolutely infeasible. Two remark-able examples are SpineAssist’s GO-LIF (20) and thesub-arachnoid space exploration permitted by MINOSC(9). These projects are examples that robotic surgery cannot only improve existing interventions, they can also be‘enabling’ technology (20).

However, the field of robotics for spinal surgery still facesconsiderable challenges. Sub-millimetre precision in instru-ment placement, under realistic conditions, has not yetbeen achieved. Patient tracking and registration still havedemanding problems, as no technology is capable of simul-taneously offering high accuracy, low invasiveness, lowradiation and high robustness. Also, the relative motion ofvertebrae introduces an additional problem in registration,which still needs to be addressed properly. In economicterms, ingenious solutions are needed to bring down thecost of robotic systems. The latter is still high and hasprevented a more widespread use of surgical robots, ashospitals are unsure whether their clinical benefitsoutweigh their elevated price. All these present challengesshow that spinal surgical robotics still has potential thatcan be transformed into increased patient well-being.

Acknowledgements

This review was financially supported by the Asociaciónde Amigos, University of Navarra, and the Ministry ofScience and Innovation, Government of Spain.

Conflict of Interest

The authors have stated explicitly that there are no conflictsof interest in connection with this article.

Funding

No specific funding.

References1. Stoianovici D, Cleary K, Patriciu A, et al. Acubot: a robot

for radiological interventions. IEEE Trans Rob Autom 2003;19(5): 927–930.

2. Melzer A, Gutmann B, Remmele T, et al. INNOMOTION forpercutaneous image-guided interventions: principles and evaluationof this MR- and CT-compatible robotic system. IEEE Eng MedBiol Mag 2008; 27(3): 66–73.

3. Tovar-Arriaga S, Tita R, Pedraza-Ortega JC, et al. Developmentof a robotic FD-CT-guided navigation system for needleplacement-preliminary accuracy tests. Int J Med Robot 2011;7(2): 225–236.

4. Onogi S, Morimoto K, Sakuma I, et al. Development of theneedle insertion robot for percutaneous vertebroplasy. InMedical Image Computing and Computer-Assisted Intervention– MICCAI 2005, Duncan JS, Gerig G (eds.). Springer: Berlin,Heidelberg, 2005; 105–113.

5. Cleary K, Watson V, Lindisch D, et al. Precision placement ofinstruments for minimally invasive procedures using a ‘needledriver’ robot. Int J Med Robot 2005; 1(2): 40–47.

6. Romanelli P, Adler JR. Technology Insight: image-guided roboticradiosurgery – a new approach for noninvasive ablation of spinallesions. Nat Clin Pract Oncol 2008; 5(7): 405–414.

7. Yang MS, Jung JH, Kim JM, et al. Current and future of spinalrobot surgery. Korean J Spine 2010; 7(2): 61–65.

8. Kantelhardt SR, Martinez R, Baerwinkel S, et al. Perioperativecourse and accuracy of screw positioning in conventional, openrobotic-guided and percutaneous robotic-guided, pedicle screwplacement. Eur Spine J 2011; 20(6): 860–868.

9. Ascari L, Stefanini C, Bertocchi U, et al. Robot-assistedendoscopic exploration of the spinal cord. Proceedings of theInstitution of Mechanical Engineers, Part C. J Mech Eng Sci2010; 224(7): 1515–1529.

10. Sautot P, Cinquin P, Lavallee S, et al. Computer assisted spinesurgery: a first step toward clinical application in orthopaedics.In Proceedings of the Annual International Conference of theIEEE, Engineering in Medicine and Biology Society. Paris,France, 1992; 1071–1072.

11. Santos-Munné JJ, Peshkin MA, Stulberg SD, et al. A stereotactic/robotic system for pedicle screw placement. In Proceedings ofthe Medicine Meets Virtual Reality III Conference, Morgan K,Satava R, Sieburg H, et al. (eds.). IOS Press/Ohmsha: San Diego,CA, 1995; 326–333.

12. Niesing B. Robots for spine surgery. Fraunhofer Mag 2001:46–47.

13. Gomes P. Surgical robotics: Reviewing the past, analysing thepresent, imagining the future. Robot Comput Integr Manuf2010: 1–6.

14. Shoham M, Burman M, Zehavi E, et al. Bone-mounted miniaturerobot for surgical procedures: concept and clinical applications.IEEE Trans Rob Autom 2003; 19(5): 893–901.

15. Hadomi O. Robotic spine surgery by Mazor Robotics, 2012 [cited30 April 2012]; http://www.mazorrobotics.com/images/pdf/mazorrobotics_company_profile.pdf

16. Sukovich W, Brink-Danan S, Hardenbrook M. Miniaturerobotic guidance for pedicle screw placement in posteriorspinal fusion: early clinical experience with the SpineAssist.Int J Med Robot 2006; 2(2): 114–122.

17. Lieberman IH, Togawa D, Kayanja MM, et al. Bone-mountedminiature robotic guidance for pedicle screw and translaminarfacet screw placement: part I –technical development and a testcase result. Neurosurgery 2006; 59(3): 641–650.

18. Togawa D, Kayanja MM, Reinhardt MK, et al. Bone-mountedminiature robotic guidance for pedicle screw and translaminarfacet screw placement: part 2 –evaluation of system accuracy.Neurosurgery 2007; 60(2, suppl 1): ONS129–139.

19. Devito DP, Kaplan L, Dietl R, et al. Clinical acceptance and accuracyassessment of spinal implants guided with SpineAssist surgicalrobot: retrospective study. Spine 2010; 35(24): 2109–2115.

20. Shoham M. Robotic Surgery: Enabling technology? In SurgicalRobotics, Rosen J, Hannaford B, Satava RM (eds.). Springer:Boston, MA, 2011; 247–255.

21. Chung GB, Kim S, Lee SG, et al. An image-guided roboticsurgery system for spinal fusion. Int J Control Autom Syst2006; 4(1): 30–41.

22. Lee J, Hwang I, Kim K, et al. Cooperative robotic assistant withdrill-by-wire end-effector for spinal fusion surgery. Indust RobotInt J 2009; 36(1): 60–72.

23. Kim S, Chung J, Yi B-J, et al. An assistive image-guidedsurgical robot system using O-arm fluoroscopy for pedicle screwinsertion: preliminary and cadaveric study. Neurosurgery 2010;67(6): 1757–1767.

24. Ortmaier T, Weiss H, Hagn U, et al. A hands-on robot foraccurate placement of pedicle screws. In Proceedings, IEEEInternational Conference on Robotics and Automation, 2006(ICRA 2006). Orlando, Florida, 2006; 4179–4186.



http://www.mazorrobotics.com/images/pdf/mazorrobotics_company_profile.pdf

http://www.mazorrobotics.com/images/pdf/mazorrobotics_company_profile.pdf

25. Ortmaier T, Weiss H, Döbele S, et al. Experiments on robot-assisted navigated drilling and milling of bones for pedicle screwplacement. Int J Med Robot 2006; 2(4): 350–363.

26. Kostrzewski S, Duff JM, Baur C, et al. Robotic system for cervicalspine surgery. Int J Med Robot 2012; 8(2): 184–190.

27. Boschetti G, Rosati G, Rossi A. A haptic system for robotic assistedspine surgery. Proceedings of IEEE Conference on Control Applica-tions, 2005 (CCA 2005); 19–24.

28. Rosati G, Rossi A, Boschetti G, et al. First experimental results ofan integrated robotic system for haptic teleoperation. In IEEEInternational Symposium on Industrial Electronics, 2007. Vigo,Spain, 2007; 3138–3143.

29. Jin H, Wang L, Hu Y, et al. Design and control strategy of roboticspinal surgical system. In IEEE/ICME International Conference onComplexMedical Engineering, 2011.Harbin, China, 2011; 627–632.

30. JinH,HuY, Li F, et al. Safety design and control algorithm for roboticspinal surgical system. In First International Conference on Robot,Vision and Signal Processing, 2011. IEEE: Shenzhen, China, 2011;190–194.

31. Wang T, Luan S, Hu L, et al. Force-based control of a compactspinal milling robot. Int J Med Robot 2010; 6(2): 178–185.

32. Cleary K, Clifford M, Stoianovici D, et al. Technology improve-ments for image-guided and minimally invasive spine procedures.IEEE Trans Inf Technol Biomed 2002; 6(4): 249–261.

33. Badaan S, Petrisor D, Kim C, et al. Does needle rotation improvelesion targeting? Int J Med Robot 2011; 7(2): 138–147.

34. Hempel E, Fischer H, Gumb L, et al. An MRI-compatible surgicalrobot for precise radiological interventions. Comput Aided Surg2003; 8(4): 180–191.

35. Moche M, Zajonz D, Kahn T, et al. MRI-guided procedures invarious regions of the body using a robotic assistance system ina closed-bore scanner: preliminary clinical experience andlimitations. J Magn Reson Imaging 2010; 31(4): 964–974.

36. Onogi S, Gotoh M, Nakajima Y, et al. Vertebral robotic puncturefor minimally invasive spinal surgery: puncture accuracy evalua-tion for vertebral model. Int J Comput Assist Radiol Surg 2009;4(suppl 1): 121–122.

37. Ju H, Zhang J, An G, et al. A robot-assisted system for mini-mally invasive spine surgery of percutaneous vertebroplasty

based on CT images. In IEEE Conference on Robotics, Auto-mation and Mechatronics, 2008. Chengdu, China, 2008;290–295.

38. Song Y, An G, Zhang J. Positioning accuracy of a medical roboticsystem for spine surgery. In 2nd International Conference onBiomedical Engineering and Informatics, 2009. Tianjin, China,2009; 1–5.

39. Karas CS, Chiocca EA. Neurosurgical robotics: a review of brainand spine applications. J Robot Surg 2007; 1(1): 39–43.

40. Lee JYK, O’Malley BW, Newman JG, et al. Transoral roboticsurgery of craniocervical junction and atlantoaxial spine: acadaveric study. J Neurosurg Spine 2010; 12(1): 13–18.

41. Ponnusamy K, Chewning S, Mohr C. Robotic approaches to theposterior spine. Spine 2009; 34(19): 2104–2109.

42. Kim MJ, Ha Y, Yang MS, et al. Robot-assisted anterior lumbarinterbody fusion (ALIF) using retroperitoneal approach. ActaNeurochir (Wien) 2010; 152(4): 675–679.

43. Gerszten PC, Burton SA, Ozhasoglu C, et al. Radiosurgery forspinal metastases: clinical experience in 500 cases from a singleinstitution. Spine 2007; 32(2): 193–199.

44. Hamilton AJ, Lulu BA, Fosmire H, et al. Preliminary clinicalexperience with linear accelerator-based spinal stereotacticradiosurgery. Neurosurgery 1995; 36(2): 311–319.

45. Yu C, MainW, Taylor D, et al. An anthropomorphic phantom studyof the accuracy of Cyberknife spinal radiosurgery. Neurosurgery2004; 55(5): 1138–1149.

46. Muacevic A, Staehler M, Drexler C, et al. Technical description,phantom accuracy, and clinical feasibility for fiducial-free frame-less real-time image-guided spinal radiosurgery. J NeurosurgSpine 2006; 5(4): 303–312.

47. Ho AK, Fu D, Cotrutz C, et al. A study of the accuracy of cyberknifespinal radiosurgery using skeletal structure tracking. Neurosurgery2007; 60(2, suppl 1): ONS147–156.

48. Rampersaud YR, Simon DA, Foley KT. Accuracy requirements forimage-guided spinal pedicle screw placement. Spine 2001;26(4): 352–359.

49. Tjardes T, Shafizadeh S, Rixen D, et al. Image-guided spinesurgery: state of the art and future directions. Eur Spine J2010; 19(1): 25–45.

A. Bertelsen et al.


a review of surgical robots for spinal interventions - … multiple electronic databases were...

Documents