Workpackages

There are four workpackages in the FAST-PATH project:

Workpackage 1 - Development and Application of Automated Pattern Recognition Tools for Discrimination of Morphological Subtypes in Prostate Cancer

This work package will develop a computer-based image analysis system for the automated assessment and recognition of pathological subtypes in prostatic tissue. We propose to combine, using sophisticated supervised and non-supervised machine learning approaches, the texture and pattern recognition approaches developed by the academic partners, UCD and QUB. This system will use a combination of knowledge-guided image processing, texture analysis and sophisticated pattern analysis to identify key tissue features on whole slide scans and automatically segment tissue samples into regions showing different morphological subtypes. This development will be carried out using tissue samples from UCD and QUB. Transferring this novel approach into an established tissue imaging platform application will be facilitated by iPATH, with independent patient cohorts available from LUND being used to validate the system. This application could have major implications for (i) improving routine diagnostic interpretation and grading of prostate resections and needle aspirates, (ii) the identification of new morphological subtypes that improve on traditional Gleason grading, and (iii) providing a high-throughput imaging framework for pathological subtyping prior to biomarker imaging at a cellular level, outlined in subsequent work packages.

Workpackage 2 - Development and Application of Automated Image Analysis Approaches for Quantitation of Prognostic Biomarkers in Prostate Cancer

This work package will initially address the problem of prostate cancer-specific pattern recognition by completing the pattern recognition functionalities of a pre-existing approach developed within UCD26, which has been recently licensed to ONCO. Additional pattern recognition functionalities will be added and tailored specifically for histopathological sub-patterns within prostate tissue. Both UCD and QUB have prior expertise in this area and existing state-of-the-art will be used to include these additional pattern recognition functionalities. In addition, prostate pattern recognition methods developed and validated in WP1 will also be integrated as independent analysis modules for biomarker validation in prostate cancer. This development will be carried out by ONCO using tissue samples and candidate prognostic markers from UCD and QUB (e.g. PRDX1, SFRP2). The developed method will then be independently validated for specificity and sensitivity in samples and candidate markers from LUND (e.g. AR, MSMB). Finally, set of markers from each site (i.e. UCD, QUB, LUND) will be evaluated in order to develop an automated prognostic panel for prostate cancer. The preferred outcome of this new IHC panel would be improved prognostic sensitivity and specificity for prostate cancer compared to existing biomarkers.

Workpackage 3 - Design and Development of High-Performance Computing Solutions for the High-throughput Analysis of Prostate Cancer Samples

The primary objective for this work package will be to explore a range of HPC platforms to determine the best solution for the specific purpose of processing prostate tissue samples. The following platforms will be investigated at QUB: (1) distributed computing using a high performance HP cluster server (>9,000 processor cores), (2) distributed Grid computing using NGS (National Grid Services), (3) a dedicated nVidia Tesla personal supercomputer approach with the use of next-gen FERMI Graphics Processing Unit (GPU) (448 CUDA cores per GPU board), and (4) a hybrid distributed solution using the high-performance cluster at QUB with the integration of Tesla S2050 1U GPU racks. The QUB team will design and develop optimised parallel processing software architecture followed by performance evaluation of these solutions, aimed at high-throughput multiplex analysis of biomarkers on prostate TMAs. This will be interfacing with the iPATH product, PathXL, for the rapid processing and evaluation of dedicated tumour identification (WP1) and biomarker analytic (WP2) algorithms. This represents the core essence of the FAST-PATH project, allowing the rapid development of novel biomarker companion algorithms for prostate cancer and their high-throughput clinical evaluation on large numbers of tissue samples.

Workpackage 4 - Implementation of Online Content-Based Image Retrieval, Annotation and Quantitation System

As virtual slide libraries begin to escalate in size, there will be an increasing need to build robust image search capabilities. This should not be restricted to the searching of text-based clinical and pathologic metadata associated with the image, but should also be based on the content of the images themselves. Content-based image retrieval (CBIR) is becoming increasingly popular in search engines such as Google, although it has never been successfully developed as a commercial application in digital pathology. The challenge here will be the large size of images concerned and the range of information that the image contains, e.g. at multiple resolutions. In this work package, we aim to build the first CBIR system for whole slide imaging in pathology. This will initially focus on prostate pathology but will be extendable to other tissue types.