High Performance Computing for Tissue MicroArray Analysis

Dr Yinhai WangDavid McCleary, Ching-Wei Wang, Jackie

James, Dean Fennell, Peter Hamilton

Introduction

Tissue Microarrays

Key technique for high throughput single assay platform for tissue biomarker research and discovery.

*Dolled-Filhart and Rimm, Principles and Practice of Oncology, 7th Edition, Chapter 7, 2004.

232 Tissue Cores

The Bottleneck

Relies on visual scoring of tissue biomarkers by pathologists. It is time consuming, subjective and prone to error.

103,790×58,586 pixels17GB

Image Analysis of TMA Virtual Slides

A TMA virtual slide is an ultra-large digital image, scanned at a high magnification (40X).

Computer assisted analysis using TMA virtual slides.Objective and reproducible.Speed?

Objective

Objective

Automate TMA analysisGenuine high throughput

platformReduce pathologists workloadSpeedup biomarker discovery

Materials and Methods

Hewlett-Packard Blade Server. Intel Xeon quad-core x86_64

processors.>9,000 processor-cores

available. 10-16GB memory per node (8

cores).Gigabit Ethernet connection.Fibre connection to hard disks

(SAN).

High Performance Computing (HPC) Platform

Parallel Processing Module

Glass slide

Image generation

Digital SlideServing Module

Image File Access Module

Analytical Module

TMA Database

Viewing

Instructions

Results

High resolution image

HPC Platform Visualisation

High Performance Computing

Colour conversi

on

Vendor format independentBGR BGRRegion

extraction Raw image

JPEG decoder

Compressed?uncompressed data

RGB GB

Pixel (n,0) Pixel (0,1)

Hamamatsuvirtual slide

No

Yes

Regionextraction Raw image

JPEG decoder

Compressed?Aperio

virtual slideNo

Yes

uncompressed dataB GR G R

Pixel (0,0) Pixel (0,1)

a

Format?

JPEG 2000decoder

JPEG 2000

JPEG

Regionextraction Raw image

JPEG decoder

Compressed?Carl Zeiss

virtual slideNo

Yes

uncompressed dataRGB RG

Pixel (0,0) Pixel (0,1)

B Colour conversi

on

Colour conversi

on

HPC: Image File Access Module

Master Worker 1 Worker 2 Worker 3 Worker 4Database

A1 B1 C1 D1

A2 B2 C2 D2

Request for corecoordinates

TMA core location (x, y) at TMA virtual slide

Assign toavailable workers

1

2 3Storage

Locate Image in Storage and Core Sub-image

4

Retrieve and LoadCore Sub-image

5

Analyse and return results

6

7Analyse and

return results

6

Informs Master itis now available

Centralised Dynamic Load Balancing

HPC: Parallel Processing Module

HPC: Analytic Module

Texture feature calculation◦ Tumour pattern recognition◦ Tumour region identification

HPC: Analytic Module

Automated quantisation of biomarker IHC density on TMA core images, using colour decomposition.

www.pathxl.com

HPC: Digital Slide Serving Module

Results

•106,290×65,017 pixels•19.3GB•229 Tissue Cores Speedup=(Fastest Sequential Code)/(Parallel

Code)=42.58

Texture Pattern Calculation for TMA Slides

Time for Loading vs. Saving

Time for actual Texture Feature Calculation

Loading, Storing, Texture

Processing time: 30minutes77secondsSpeedup=22.19

Biomarker Quantification

Multi TMA Slides

There are >9000 processor-cores available

The processing of 1 TMA virtual slide uses <100 processor-cores.

>90 TMA virtual slides can be processed simultaneously (≈1 minute).

Genuine high throughput platform for multiplex multi-TMA studies.

Conclusion• A novel high performance computing

platform for the rapid analysis of TMA virtual slides.

• The centralised load balancing approach is proven to be robust.

• It significantly speedups up the analysis of TMAs, removing the bottleneck.

• Valuable platform for TMA research & biomarker discovery.

• High performance platform for the algorithm prototyping, development & evaluation.

Acknowledgements

Thanks

Dr Yinhai Wangy.wang@qub.ac.uk0044-(0)-28-9097 5816