Scandinavian Journal of Gastroenterology. 2013; 48: 1302–1307

ORIGINAL ARTICLE

Visualization of colorectal neoplasia by a second-generationautofluorescence imaging system

DAISUKE IDE1, NAOTO TAMAI2, HIROKO INOMATA2, TOMOHIKO R OHYA2,HIROYUKI AIHARA2, SYOICHI SAITO2, TOMOHIRO KATO2 & HISAO TAJIRI1, 2

1Department of Internal Medicine, Division of Gastroenterology and Hepatology, The Jikei University School of Medicine,Tokyo, Japan, and 2Department of Endoscopy, The Jikei University School of Medicine, Tokyo, Japan

AbstractObjective. Autofluorescence imaging (AFI) systems may allow better visualization of colorectal neoplasia than conventionalmethods. However, this is difficult to demonstrate objectively. Recently, a second-generation AFI system with a noise-reduction algorithm was developed. We aimed to objectively evaluate the visualization of colorectal neoplasia by using asecond-generation AFI system and software to calculate the color-contrast index.Material andmethods.We retrospectivelyreviewed 53 consecutive colorectal neoplasias examined using the second-generation AFI system. Color-contrast indicesbetween the colorectal lesions and the surrounding normal mucosa in the WLI, AFI and NBI images were calculated. TheWLI, AFI, NBI and CE images were also evaluated by endoscopists using questionnaire-based visualization scores. Results.The color-contrast index seen in the AFI images (33.74 ± 9.20) was significantly higher than that in either the WLI (11.14 ±6.14) or NBI images (11.72 ± 7.12). There was no significant difference between the color-contrast indices of the WLI andNBI images. The mean AFI image visualization score (6.7 ± 1.8) was significantly higher than that of WLI (6.0 ± 1.7), andtended to be higher than that of the NBI images (6.1 ± 1.6) when assessed by less-experienced endoscopists. Conclusions.This study objectively demonstrates that compared to WLI and NBI, the second-generation AFI system enables superiorvisualization of colorectal neoplasms. The visualization scores were higher for the AFI images when evaluated by less-experienced endoscopists. These results indicate that the second-generation AFI systemmay aid less-experienced endoscopistsin the detection of colorectal neoplasia.

Key Words: autofluorescence imaging, colon, colorectal neoplasia, narrow-band imaging

Introduction

Colorectal neoplasia is one of the most commoncancers worldwide, and its prevalence is steadilyincreasing in Japan [1]. Early detection and appro-priate removal of colorectal neoplasia is necessary toreduce the mortality associated with this condition [2–4]. Colonoscopy is considered to be the most effectiveprocedure for both direct visualization and in situtreatment of the lesions. However, according to aprevious report, the miss rate for adenomas ‡1 cmwas 6%, and small or flat neoplasia may be missedmore frequently [5]. These findings highlight the need

for improvements in neoplasia detection duringcolonoscopy.If colorectal neoplasia is missed on colonoscopy,

surgery is more likely to be needed and the mortalityrate is likely to increase [6]. In order to improvedetection rates of colorectal neoplasia, the develop-ment of a new endoscopic modality is required.Evidence regarding the use of autofluorescence

imaging (AFI) in the detection of colorectal neoplasiais conflicting [7,8]; however, several reports concludethat AFI can improve visualization of colorectal neo-plasia [9,10]. One study reported better visualization ofcolorectal neoplasia with AFI than with white-light

Correspondence: Naoto Tamai, MD PhD, Department of Endoscopy, The Jikei University School of Medicine, 3-25-8 Nishi Shinbashi, Minato-ku, Tokyo105-8461, Japan. Tel: +81 3 3433 1111 (Ext. 3181). Fax: +81 3 3459 4524. E-mail: tamai-naoto@jikei.ac.jp

(Received 29 May 2013; revised 4 August 2013; accepted 7 August 2013)

ISSN 0036-5521 print/ISSN 1502-7708 online � 2013 Informa HealthcareDOI: 10.3109/00365521.2013.834073

Scan

d J

Gas

troe

nter

ol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.

imaging (WLI) or narrow-band imaging (NBI); how-ever, only laterally spreading tumors (LSTs) were con-sidered and conclusions were based on questionnaires[10]. These mixed results suggest that a more thoroughassessment of the capabilities of AFI is needed.Recently, a second-generation AFI system com-

posed of light sources (EVIS LUCERA ELITECLV-290SL; Olympus Medical Systems, Tokyo,Japan) and video processors (EVIS LUCERA ELITECV-290; Olympus Medical Systems) was developed.The first-generation AFI systems (EVIS LUCERACLV-260SL, EVIS LUCERA CV-260S; OlympusMedical Systems) had technical problems includingflickering and color splitting in the endoscopic imagebecause of a slower frame rate than in WLI. There aretwo main differences between the first- and second-generation AFI systems. The first major improvementis a brighter lamp that allows for a higher frame rate,even though the overall brightness of the endoscopicimages is equal to that from first-generation AFIsystems. Therefore, there is less flickering and colorsplitting in the endoscopic images. The second majorimprovement is in the image processing algorithm,specifically the noise reduction algorithm, whichresults in higher resolution images with less noiseinterference.In addition, to minimize the subjectivity of image

analysis, our group developed color analysis softwarein collaboration with Olympus Medical Systems thatautomatically calculates the color-contrast index ofthe images.The aim of this study is to evaluate the visualization of

colorectal neoplasia (including protruded lesions)observed using a second-generationAFI system througha comparison of the conventional questionnaire-basedapproach with an objective approach using color-contrast index software.

Methods

We retrospectively reviewed the records from 53 con-secutive patients who underwent colonoscopy using

the second-generation AFI system with a colonoscopedesigned for AFI observation (CF-FH260AZI; Olym-pus Medical Systems) before endoscopic or surgicaltreatment for colorectal neoplasia at the Jikei Univer-sity Hospital from May 2011 to July 2012.All patients were given a polyethylene glycol-

electrolyte solution to empty the intestine for colonos-copy. When bowel preparation was deemed adequate,a survey endoscopy using the WLI function of the AFIsystem was performed by experienced endoscopists.Detected colorectal lesions were then observed withAFI and NBI modes by switching from WLI to AFIand NBI with a button on the control head of theendoscope. Subsequently, the lesions were observed bychromoendoscopy (CE) using 0.4% indigo carminewith the white-light mode (Figure 1). The four endo-scopic images of the colorectal lesions were automat-ically saved in the electronic medical records. Anexperienced endoscopist, blinded to this study,selected the most representative of each of the imageformats (WLI, AFI, NBI and CE) for each colorectalneoplasia.

Color-contrast index

Endoscopic images were automatically saved in an8-bit bitmap format on a computer linked to thevideo-endoscopy system. The false color range inthe AFI system was decided based on the balanceof the autofluorescence intensity and reflected greenlight intensity. The thickness of the lesion, degree ofvascularity and glandular density could affect thisbalance. The red, green and blue intensities in thestandard RGB color tone were monitored with thecolor analysis software that was developed in conjunc-tion with Olympus Medical Systems Corp., Tokyo,Japan. This software analyzes the R, G and B valuesand their inverse gamma correction values for thesample areas. The size of the region of interest(ROI) for color-tone sampling was a 10 � 10-pixelsquare. To describe the color-contrast between thelesion and the surrounding normal mucosa, we used

A B C D

Figure 1. Images of colorectal neoplasia observed by white-light imaging (A), autofluorescence imaging (B), narrow-band imaging (C), andchromoendoscopy (D).

Visualization of colorectal neoplasia 1303

Scan

d J

Gas

troe

nter

ol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.

the DE*ab value in the L*a*b* color space which wasdefined by the Commission Internationale del’Eclai-rage (CIE) as a color-contrast index. In the L*a*b*color space, values were shown with two color axes(a* axis and b* axis) and a single brightness axis(L* axis). The DE*ab value was obtained by calcu-lating the Euclidean distance between the twocoordinates in the L* a* b* color space, which corre-sponded to the color difference perceived by theendoscopists. The conversion equations of the DE*abvalue are described in Figure 2. Color-tone samplingwas performed from three areas in the lesions andthree areas in the surrounding normal regions(Figure 3), and the DE*ab values of the WLI, AFIand NBI images were calculated on the basis of themean R, G and B values of the three areas (lesion/surrounding normal mucosa).

The visualization score

The images of all the lesions were randomly arranged,and aMicrosoft PowerPoint presentation was created.These images did not contain any information toidentify the patient or the lesion. The PowerPointpresentations were sent to the respective raters fortheir independent evaluation. The WLI, AFI, NBIand CE images were evaluated by endoscopists usingquestionnaire-based visualization scores. The visual-ization score was defined as follows: the worst visu-alization was scored as 0 and the best as 10 (scoreswere measured as continuous variables using thevisual analogue method). The images were assessedby two groups of endoscopists (A and B). Group Acomprised three endoscopists with no previous expe-rience in image-enhanced endoscopy (IEE), andGroup B comprised three endoscopists, who hadeach analyzed over 500 cases using IEE. Each endo-scopic image was assessed and given a visualizationscore based on two criteria: (1) the ability to detect thelesion and (2) the clarity of the tumor margins. Toevaluate the visualization of colorectal neoplasia, we

calculated the average visual analogue scale scores forWLI, AFI, NBI and CE in both groups.

Statistical analysis

Statistical analysis was performed using SPSS for Win-dows (SPSS, release 6.0, 1993; SPSS Inc., Chicago,Illinois, USA). Data are expressed as mean ± SD. Todetermine differences in the mean color-contrastindices (DE*ab values) and visualization scores, com-parisons between the groups were performed byone-way analysis of variance (ANOVA), followed bymultiple comparison testing using the Bonferroni–Dunn method. Continuous variables were analyzedusing a t-test. A p Value of < 0.05 was consideredstatistically significant and a p Value of < 0.1 wasinterpreted as a tendency.

X = 0.4124R + 0.3576G + 0.1805BY = 0.2126R + 0.7152G + 0.0722BZ = 0.0193R + 0.1192G + 0.9505B

L* = 116(Y/Yn)-16a* = 500[(X/Xn)-(Y/Yn)]b* = 200[(Y/Yn)-(Z/Zn)]where Xn = 0.9505, Yn = 1.0000, and Zn =1.0890

When coordinates of two color tones in L*a*b* color space are indicated as (L*1, a*1, b*1) and(L*0, a*0, b*0), respectively: ΔE*ab = {(ΔL*)2 + (Δa*)2 + (Δb*)2}1/2

where ΔL* = L*1 – L*0, Δa* = a*1 – a*0, and Δb* = *b1 – b*0

Equations for conversion from analyzed R (red),G (green) and (blue) values in sRGB colorspace to CIE XYZ color space

Equations for conversion from CIE XYZ colorspace to CIE 1976 L*a*b* color space

Figure 2. The conversion equation as defined by the Commission Internationale del’Eclairage.

Figure 3. An image of colorectal neoplasia acquired using auto-fluorescence imaging: color-tone sampling was performed fromthree areas of the lesions and the surrounding normal regions.Sampling areas of the lesion are indicated by yellow squares, andthat of the surrounding normal regions by red squares.

1304 D. Ide et al.

Scan

d J

Gas

troe

nter

ol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.

Results

A total of 53 patients (53 lesions) were retrospectivelyreviewed in this study. The mean patient age was64.9 ± 10 years, the male to female ratio was 1:1.52,and the mean lesion size was 36.3 ± 12.3 mm. His-tologically, of the 53 lesions, there were 10 tubularadenomas (18.9%), 29 mucosal cancers (54.7%) and14 submucosal or deeper cancers (26.4%). Macro-scopic types included 25 protruded (47.2%) and28 flat elevated (52.8%). Tumor location included7 in the cecum (13.2%), 30 in the right colon(56.6%), 8 in the left colon (15.1%) and 8 in therectum (15.1%) (Table I).The color-contrast indices calculated for the dif-

ferently acquired images of colorectal neoplasiashowed significant differences in DE*ab values forWLI (11.14 ± 6.14), AFI (33.74 ± 9.20), and NBI(11.72 ± 7.12) (p < 0.0001). The DE*ab values of AFIimages were significantly higher than those for WLIand NBI images (p < 0.001). However, there was nosignificant difference in DE*ab values between WLIand NBI (Figure 4).There was no significant difference in the average

visualization scores between WLI (6.9 ± 2.0), AFI(6.9 ± 1.9), NBI (6.8 ± 1.8) and CE (7.1 ± 1.6) asrated by Group B. In contrast, the average visualiza-tion score of AFI images (6.7 ± 1.8) was significantlyhigher than that of WLI images (6.0 ± 1.7) (p < 0.05),and tended to be higher than that of NBI images(6.1 ± 1.6) (p < 0.1), when assessed by group A. Therewas no significant difference in average visualizationscores between AFI and CE in this group (Table II).In AFI, the DE*ab values and the visualization scoresof protruded lesions (36.99 ± 7.56) (5.9 ± 0.8) were

significantly higher than those of flat elevated lesions(30.84 ± 9.90) (5.4 ± 0.8) (Table III).

Discussion

Colonoscopy is considered the gold standard for thedetection and treatment of colorectal neoplasia. Earlydetection and removal of colorectal neoplasia isconsidered the most effective way of preventing pro-gression [2,11]. The conventional colonoscopic tech-nique during withdrawal, even if performed carefully,cannot detect all lesions. This limitation has led to thedevelopment of alternative techniques aimed atimproving the detection rate of colorectal neoplasia.For example, the use of CE can improve detectionrates of colorectal neoplasia; specifically, the use ofpan-CE improves the detection rate of diminutive andflat neoplasias [12–14]. However, in these studies, thewithdrawal time for the indigo carmine dye spraygroup was twice as long when compared to the controlgroup [15]. This finding illustrates the challengesassociated with this procedure.It has been shown that NBI colonoscopy improves

the detection of colorectal neoplasia [16]. However, ithas also been reported that when compared to con-ventional WLI, NBI did not improve the detection

Table I. Summary of case characteristics.

Number of patients 53Number of lesions 53Age, years, means 64.9 ± 10.7Sex, male/female 21/32Size of the lesion, mm, meanlon 36.3 ± 12.3Histology of the lesionTubular adenoma/Mucosalcancer/Submucosal or deeper cancers

10/29/14

Macroscopic type of the lesionsProtruded/Flat elevated 25/28Location of the lesionCecum/Right colon/Left colon/Rectum 7/30/8/8

WLI(n = 53)

mean 11.14 ± 6.14

NBI(n = 53)

mean 11.72 ± 7.12

AFI(n = 53)

mean 33.74 ± 9.20

ΔE*a

b v

alu

e

0

10

20

30

40

50

60

* **

Figure 4. The distribution of color-contrast indices in colorectalneoplasia for each modality. Data are shown as the mean (SD). *,**p < 0.001, one-way ANOVA and post hoc tests (Bonferroni/Dunnmethod).

Table II. The visualization scores of each modality.

White-light imaging Autofluorescence imaging Narrow-band imaging Chromoendoscopy p Value*

Group A 6.0 ± 1.7 6.7 ± 1.8 6.1 ± 1.6 6.3 ± 1.6 0.02Group B 6.9 ± 2.0 6.9 ± 1.9 6.8 ± 1.8 7.1 ± 1.6 0.55

*One-way ANOVA.

Visualization of colorectal neoplasia 1305

Scan

d J

Gas

troe

nter

ol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.

rate of colorectal neoplasia in a multicenter random-ized controlled trial [17]. To date, therefore, evidenceas to the advantages of NBI for the detection ofcolorectal neoplasia is still contradictory.A pilot study using AFI demonstrated the superi-

ority of this technique compared with WLI in tumordetection [7]. In a randomized controlled trial, thesensitivity of AFI was significantly higher than that ofWLI for the detection of colorectal neoplasia in high-risk patients from families with Lynch syndrome orfamilial CRC [18]. However, conflicting results havebeen reported elsewhere [8].This is the first study to assess the visualization of

colorectal neoplasia objectively using software to cal-culate a color-contrast index (DE*ab value). In addi-tion, as far as we know, this is the first reportevaluating a second-generation AFI system (EVISLUCERA ELITE CLV-290SL, EVIS LUCERAELITE CV-290; Olympus Medical Systems). Theseresults show that AFI provides a significantly highercolor-contrast index than WLI and NBI, suggestingthat visualization of colorectal neoplasia, with regardto lesion detection and margin clarity, is better inimages acquired in this manner. Accordingly, theaverage visualization score of AFI was significantlyhigher than that of WLI, and tended to be higher thanthat of NBI when assessed by less-experiencedendoscopists. However, when images were assessedby experienced endoscopists, there was no statisticallysignificant difference in the average visualizationscores of images made using WLI, AFI, NBI orCE. This may reflect the fact that these endoscopistshad more experience with WLI and NBI images thanwith images acquired with AFI, which is a relativelynew diagnostic tool in Japan. In this study, we alsocompared DE*ab values and visualization scores ofAFI images between protruded lesions and flat ele-vated lesions. The DE*ab values and visualizationscores of protruded lesions were significantly higherthan those of flat elevated lesions. This result mayindicate that the AFI system enables visualization ofprotruded lesions more clearly than flat elevatedlesions. There are several limitations in this study:the mean lesion size in this study was 36.3 mm, andthe majority of lesions were located in the right colon.The larger lesion sizes and location of the lesionsshould be considered biases of this study, though

consecutive lesions were evaluated. Furthermore,the color-contrast index and visualization scorewere only evaluated in still images in this study.Therefore, it cannot be concluded that AFI improvesthe detection of colorectal neoplasia during colonos-copy and further study using video clips to evaluatethe second-generation AFI system for detectability ofcolorectal lesions is needed. However, the betterlesion visualization that is possible in images acquiredwith AFI might lead to an overall improvement indetection rate. A randomized, prospective study com-paring the detection rate of colorectal neoplasia ineach modality would be a useful follow-on to thisstudy.In conclusion, this study objectively demonstrates

that the second-generation AFI system allows super-ior visualization of colorectal neoplasms compared toWLI and NBI. Less experienced endoscopists scoredthe ease of tumor visualization as being higher in AFIimages than for either WLI or NBI images. This studyindicates that second-generation AFI systems may aidless experienced endoscopists in the detection ofcolorectal neoplasia.

Declaration of interest: The second-generationAFI system was provided free of charge by OlympusMedical Systems (Tokyo, Japan).

References

[1] Matsuda T, Marugame T, Kamo K, Katanoda K, Ajiki W,Sobue T, et al. Cancer incidence and incidence rates in Japanin 2006: Based on data from 15 population-based cancerregistries in the monitoring of cancer incidence in Japan(MCIJ) project. Jpn J Clin Oncol 2012;42:139–47.

[2] Winawer SJ, Zauber AG, Ho MN, O’Brien MJ, Gottlieb LS,Sternberg SS, et al. Prevention of colorectal cancer by colo-noscopic polypectomy. the national polyp study workgroup.N Engl J Med 1993;329:1977–81.

[3] Citarda F, Tomaselli G, Capocaccia R, Barcherini S,Crespi M; Italian Multicentre Study Group. Efficacy instandard clinical practice of colonoscopic polypectomy inreducing colorectal cancer incidence. Gut 2001;48:812–15.

[4] Newcomb PA, Storer BE, Morimoto LM, Templeton A,Potter JD. Long-term efficacy of sigmoidoscopy in the reduc-tion of colorectal cancer incidence. J Natl Cancer Inst 2003;95:622–5.

[5] Rex DK, Cutler CS, Lemmel GT, Rahmani EY, Clark DW,Helper DJ, et al. Colonoscopic miss rates of adenomasdetermined by back-to-back colonoscopies. Gastroenterol-ogy 1997;112:24–8.

[6] Hurlstone DP, Cross SS, Slater R, Sanders DS, Brown S.Detecting diminutive colorectal lesions at colonoscopy:A randomised controlled trial of pan-colonic versus targetedchromoscopy. Gut 2004;53:376–80.

[7] Matsuda T, Saito Y, Fu KI, Uraoka T, Kobayashi N,Nakajima T, et al. Does autofluorescence imaging videoen-doscopy system improve the colonoscopic polyp detectionrate?–a pilot study. Am J Gastroenterol 2008;103:1926–32.

Table III. DE*ab values and AFI image visualization scoresbetween protruded and flat elevated lesions.

Protrudedlesions

Flat elevatedlesions p-Value

DE*ab values 36.99 ± 9.90 30.84 ± 7.56 < 0.05Visualization scores 5.9 ± 0.8 5.4 ± 0.8 < 0.05

1306 D. Ide et al.

Scan

d J

Gas

troe

nter

ol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.

[8] Kuiper T, van den Broek FJ, Naber AH, van Soest EJ,Scholten P, Mallant-Hent RC, et al. Endoscopic trimodalimaging detects colonic neoplasia as well as standard videoendoscopy. Gastroenterology 2011;140:1887–94.

[9] Suzuki H, Saito Y, Oda I, Kikuchi T, Kiriyama S,Fukunaga S. Comparison of narrowband imaging with auto-fluorescence imaging for endoscopic visualization of super-ficial squamous cell carcinoma lesions of the esophagus.Diagn Ther Endosc 2012;2012:507597.

[10] Tamai N, Saito Y, Sakamoto T, Nakajima T, Matsuda T,Vikneswaran N, et al. Visualization of laterally spreadingcolorectal tumors by using image-enhanced endoscopy. Gas-troenterol Res Pract 2012;2012:638391.

[11] Winawer SJ, Zauber AG, O’Brien MJ, Ho MN, Gottlieb L,Sternberg SS, et al. Randomized comparison of surveillanceintervals after colonoscopic removal of newly diagnosedadenomatous polyps. The National Polyp Study Workgroup.N Engl J Med 1993;328:901–6.

[12] Fujii T, Hasegawa RT, Saitoh Y, Fleischer D, Saito Y,Sano Y, et al. Chromoscopy during colonoscopy. Endoscopy2001;33:1036–41.

[13] Hurlstone DP, Cross SS, Adam I, Shorthouse AJ, Brown S,Sanders DS, et al. A prospective clinicopathological andendoscopic evaluation of flat and depressed colorectal lesions

in the United Kingdom. Am J Gastroenterol 2003;98:2543–9.

[14] Lapalus MG, Helbert T, Napoleon B, Rey JF, Houcke P,Ponchon T, et al. Does chromoendoscopy with structureenhancement improve the colonoscopic adenoma detectionrate? Endoscopy 2006;38:444–8.

[15] Brooker JC, Saunders BP, Shah SG, Thapar CJ, Thomas HJ,Atkin WS, et al. Total colonic dye-spray increases the detec-tion of diminutive adenomas during routine colonoscopy:A randomized controlled trial. Gastrointest Endosc 2002;56:333–8.

[16] Uraoka T, Saito Y, Matsuda T, Sano Y, Ikehara H,Mashimo Y, et al. Detectability of colorectal neoplasticlesions using a narrow-band imaging system: A pilot study.J Gastroenterol Hepatol 2008;23:1810–15.

[17] Ikematsu H, Saito Y, Tanaka S, Uraoka T, Sano Y,Horimatsu T, et al. The impact of narrow band imaging forcolon polyp detection: Amulticenter randomized controlled trialby tandem colonoscopy. J Gastroenterol 2012;47:1099–107.

[18] Ramsoekh D, Haringsma J, Poley JW, van Putten P,van Dekken H, Steyerberg EW, et al. A back-to-back com-parison of white light video endoscopy with autofluorescenceendoscopy for adenoma detection in high-risk subjects. Gut2010;59:785–93.

Visualization of colorectal neoplasia 1307

Scan

d J

Gas

troe

nter

ol D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y M

ichi

gan

Uni

vers

ity o

n 11

/12/

14Fo

r pe

rson

al u

se o

nly.