1

Brain Segmentation – A Case study of Biomedical

Cloud Computing for Education and Research

Victor ChangLeeds Metropolitan University (and University of Southampton)

6th June 2013, Greenwich, London

2

Overview

• Motivation

• Technology

• Benefits

• Methodology

• Results

• Discussions

• Conclusion, Questions and Answers

3

Objectives and executive summary• Medical imaging is widely adopted in healthcare. New ways to

improve medical imaging services are regularly exploited. • This paper/presentation describes the adoption of Cloud

Computing is useful for medical education and research.• A working Bioinformatics Cloud platform can demonstrate

computation and visualisation of brain imaging. • The aim is to study segmentation of brains, which divides the

brain into ten major regions.• The Cloud platform has two functions: (i) it can highlight each

region for ten different segments and (ii) it can adjust intensity of segmentation to allow basic study of brain medicine.

• Benefits include costs, saving time and costs for repeated work and making this subject easier to understand.

Introduction• Healthcare informatics has played a strategic role in the

National Health Service (NHS) and has been influential to the way in the IT project development for different NHS Trusts.

• Due to the fusion between Healthcare Cloud Computing and Education, it has become increasingly important for medical scientists and educators to use Cloud Computing.

• Other case studies are supported by case studies supported by NHS, King’s College’s London and the University of Oxford.

• There are interests for inter-disciplinary investigations between (a) learning satisfaction and brain response and (b) learning difficulties and brain response.

• The objective is to study how human brain responds while recapturing a skill that has been learned some time ago in an interactive and motivated environment offered by Cloud.

4

5

Cloud platforms and applications• Funded by NHS UK, where Guy’s and St Thomas NHS Trust

(GSTT) and King’s College London (KCL) have worked together to deliver a service. This project began in September 2008 and completed in January 2011 to serve cancer researchers.

• Cloud Storage is a PaaS, and needs careful planning and a thorough implementation. This requires using an integrated adoption of multiple vendors’ solutions.

• Bioinformatics is an area to experience rapid growth in user requirements and visualisation of complex analysis. Hence, it must be easy to use, have a good performance and able to present complex brain/medical imaging.

• Cloud Bioinformatics is a new concept and implementation. In the past, private and in-house storage is used. Maintenance of data protection and security is a challenge.

6

Cloud Technologies• Storage Area Network (SAN) can be made up of different

Network Attached Storage (NAS), so that each NAS can focus on a particular function.

• RAID 10; SCSI technology; 12 TB per SAN box; and security include SSL; additional firewall, VPN and data encryption.

• Performance improvement and monitoring: This allows tracking the overall and specific performance of the SAN cluster, and also enhances group or individual performance.

• Advanced backup: Sophisticated techniques and mechanisms (use of enterprise software is optional) are required. Services include automation, data recovery and quality of services.

• Software repository: Archive code and software development.• Bioinformatics: Completed software as a service which

integrates with the system.• Code/algorithms: C#, XML, C++, R, SQL, CUDA, MATLAB and

Mathematica; CCAF; Business Intelligence; Brain Science.Selections of technologies: Details in the next page.

7

Technology selections

What is it used Vendors involved

Focus or rationale Benefits or impacts

Network Attached Storage (NAS)

To store data and perform automated and manual/personal backup

Iomega/EMCLacieWestern DigitalHP

They have a different focus and set up. HP is more robust but more time-consuming to configure. The rest is distributed between RAID 0, 1 and 5.

Each specific function is assigned with each NAS. There are 5 NAS at GSTT/KCL site and 3 at Data Centre, including 2 for Archiving.

Infrastructure (networking and hosting solution)

Virtualisation

Collaborator and in-house

University of London Data Centre

Some services need a more secure and reliable place. University of London Data Centre offers 24/7 services with around 500 servers in place, and is ideal for hosting solution.

Amount of work is reduced for maintenance of the entire infrastructure. It stores crucial data and used for archiving, which backup historical data and backup the most important data automatically and periodically.

Backup applications

Third party and in-house

Open SourceOracleHPVmwareSymantecIn-house development

There is a mixture of in-house development and third party solution. HP software is used for high availability and reliability. The rest is to support backup in between NAS. Vmware is used for virtual storage and backup.

Some applications are good in a particular service, and it is important to identify the most suitable application for particular services.

Software Algorithms

CCAF (my own)Business IntelligenceBig Data

C#, XML, C++, R, SQL, CUDA, MATLAB, Mathematica

Develop algorithms and make the services easier to use. Process thousands of data to present results in numeric & visualisation.

Resources can be virtualised and saves effort such as replication.

Security Third party and in-house

KCL/GSTTMacafeeSymantecF5

Security is based on the in-house solution and vendor solution is focused on secure firewall and anti-virus.

Remote access is given to a list of approved users.

8

5 services

1. Statistics

2. Statistics

4. Bioinformatics

3. Database

5. Virtualisation

Initially it is used to back up digital cancer cluster. It helps backing up important data in NAS 3.

Backup and archiving

Used by Tissue Bank Group and occasionally Bioinformatics Group.

Red arrows: automated and secure backup to London University Data Centre.Blue arrows: automated, easy-to-use and secure backup internally and between GSTT and KCL.

University of London Computing (Data) Centre (ULCC) GSTT and KCL

NAS 6: Backup Database (3.63 TB at total, 1.8 TB in effective, RAID 1)

NAS 1 (3TB at total, 2.1 TB in effective, RAID 5)

NAS 2 (3TB at total, 2.1 TB in effective, RAID 5)

NAS 3 (3TB at total, 2.1 TB in effective, RAID 5)

Bioinformatics services, (IP) iscsi SAN

NAS 7 (used as a NAS): Archive (12 TB at total, 6TB effective, RAID 10)

NAS 8 (used as a NAS): Archive (12 TB at total, 6TB effective, RAID 10)

NAS 4 (3.63 TB at total: 1.8 TB in effective, RAID 1)

NAS 5 (3.63 TB at total and effective, RAID 0)

Digital cancer (2 clusters)

Bioinformatics cluster

Used by Bioinformatics Group

Used by Bioinformatics Group

Used by Epidemiology and Breast Cancer Biology Group (BCBG): This is an important Gateway NAS to backup files and archiving.

Mirror services at a different location

Architecture / Deployment

9

Selected Software as a Service: Bioinformatics

• Medical imaging, 3D insulin molecules, tumor modeling, 3D DNA and, spirals in plants and so on.

10

Some pictures

Brain segmentation (part 1)• Brain segmentation service has 2 main functions: (i) it highlights

each region for 10 different segments; and (ii) it can adjust intensity of segmentation to allow basic study of brain medicine.

• The service can be used by medical students to improve the quality of teaching and learning satisfaction amongst students.

• Two medical cohort groups with fifteen in each group. They were introduced the use of Cloud Computing for medical training, and surveys were taken before and after the adoption of Medical Cloud Computing Education (MCCE).

• Each student has used MCCE for at least three months. The focus of the study was to identify the rate of learning satisfaction before and after the introduction of MCCE. Survey questions were taken and an average score for learning satisfaction was recorded for each student.

11

Brain segmentation (part 2)• All the records are undertaken with statistical analysis presented. The variable

“group1_without_Cloud” indicates learning satisfaction for cohort group one before using MCCE and the variable “group1_with_Cloud” is the variable after using MCCE.

• The variables “group2_without_Cloud” & ““group2_with_Cloud” are the same terminology for cohort group two. Mean is the percentage for learning satisfaction before and after using MCCE. There are 20% improvements in learning satisfaction

12

Variable | Obs Mean Std. Dev. Min Max

-------------+-------------------------------------------------------------

group1_without_Cloud | 15 69.33333 3.9036 60 76

group1_with_Cloud | 15 89.4 3.601587 80 95

Variable | Obs Mean Std. Dev. Min Max

-------------+-------------------------------------------------------------

group2_without_Cloud | 15 70.06667 3.494213 65 76

group2_with_Cloud | 15 90.13333 2.899918 85 95

Statistical analysis for two medical cohort groups before and after using MCCE

Brain segmentation (part 3)• Brain segmentation aims to: 1. help scientists to understand how human brains work2. help trainee doctors to have better understanding with physiology and

biological sciences related to brain3. help scientists to make more pioneering research for brain segmentation and

the use of technologies in the aid of their analyses.• Simulations help scientists to understand how human learn new objects or

skills; and also how scientists can discover better ways of identifying tumours. • Often the detections of tumours require MRI scans to confirm, and any

pioneering ways of discovery will help scientists to improve the quality of patients’ life in the long term, and reduce the possibility of making errors in the process of treatment and operations.

• There is another initiative to study the intensity of our human brains while recapturing a skill that has not been used for two years.

• Things to consider for the chosen skill: Not too complicated; volunteers have greater fun of participating; something that can

13

Brain segmentation (part 4)• Two groups of volunteers with eight in a group were asked to dance in this

study and all of them did not dance for two years at least. Each person had attached an electronic device to record their brain activity.

• CCBST then collected all the data, calculated the average on brain segmentation and drew out the brain activities as a collective result.

• Results are presented in 3D Visualisation where the more intense the visualisation is, the more activities the brain have undertaken during the dance sessions.

• First time – done with scientific equipment. Cloud Computing – allows to reproduce the same results/impacts without repeating the work.

• 100% of volunteers are reported to have positively increased in their brain activities while recapturing their dancing skills. Results are presented in the next slide.

• The objective: To study how volunteers respond when they reuse a skill, which they learn some time ago but do not practice for ~2 years. Sophisticated equipment and expertise can be used but are expensive. Simulations can reduce the costs in a way that results can be repeated.

14

Brain segmentation (part 5)• Structure and function of human brain:

• Frontal Lobe: behaviour; intelligence; memory; movement

• Parietal Lobe: intelligence; language; reading; sensation

• Temporal Lobe: behaviour; hearing; memory; speech; vision

• Cerebellum: balance; co-ordination• Brain stem: blood pressure;

breathing; consciousness; heartbeat; swallowing

15

Brain segmentation (part 6)• Figure 1 shows the core area of the human brain, which

covers the frontal lobe, parietal lobe, occipital lobe (these three are grey matters, the last one is situated at the back), corpus callosum, hippocampus (inner brain, the later one controls emotion) and brain stem (middle brain).

• Figure 1 shows how human brain of volunteers (collective results) responds when reusing a skill such as dancing. The result shows that cells in the frontal lobe and occipital lobe are reacting actively, and the entire brain stem is responding very positively.

• This may mean volunteers require a high level of balancing. When they dance, their movement is fast, and must keep themselves balanced with swift but steady movements. The areas that represent the emotions are not so obvious but still can be seen with some activities.

16

Figure 1: The collective result of brain activities while recapturing dancing skills

Conclusion and Future Work (part 1)• Cloud Computing Brain Segmentation Technology (CCBST) and

Medical Cloud Computing Education (MCCE) are extremely useful for the healthcare scientists to make pioneering research.

• Added values:1. This improves the efficiency for teaching, as trainee doctors can

fully understand their expectations and medical lessons. There are 20% improvements in learning satisfaction and MCCE is strategic and influential to medical education.

2. Simulations can allow scientists to understand human organs such as brain segmentation. Scientists can understand how volunteers learn and their reactions to the brain, particularly the human brain activities while recapturing a skill such as dancing. 100% of volunteer participants are reported to have positive learning experience.

17

Conclusion and Future Work (part 2)

3. Cloud Computing offers cost-effective and innovative way to deliver education and research while maintaining a high level of quality of work for education, research and technology as demonstrated in these two cases.•Future Work includes the followings: 1.Investigate brain segmentation for learning a new skill2.Other bioinformatics services/applications by the use of Cloud technologies.3.Results will be published in coming conferences or journal articles (in the next 12 months).

18

19

Questions?

Thank you!!