I-Corps: Cloud pathology platform for computer aided digitized histopathology image processing and analysis system

University Of Texas At San Antonio, San Antonio TX

Investigators

Abstract

Owing to the potential negative side effects of prostate cancer treatment, there is a real need to accurately diagnose and not over treat prostate cancer. The development of a system capable of quickly and accurately processing digitized prostate biopsy slides to detect, classify and score prostate cancer will result in fewer unnecessary surgeries with potentially harmful side effects and more efficient use of healthcare dollars. This I-Corps team believes that pathologists using such a system will be more productive and better able to handle the increasing workload that an aging population will produce. The system will also provide critical access to a "digital" pathologist for developing countries where access to pathologists is severely limited. This I-Corps team will conduct customer discovery and more fully analyze the commercial potential for a cloud pathology platform for computer aided diagnostic and decision support system for prostate cancer. The team will demonstrate a system where a user will input a digitized prostate biopsy slide and the system will analyze the slide, identify cancerous tissue regions, classify and grade the prostate tumors and output a Gleason score. The proposed innovation leverages advances in computer vision, image processing and neural networks. The system utilizes a fuzzy color standardization method as a preprocessing step that allows the system to analyze images from different sources. A novel algorithm automatically segments prostate tissue structures based on color decomposition and extracts morphological and architectural features from prostate needle biopsy images and automatically classifies the prostate cancer biopsy images based on the Gleason grading system. Finally, a new high-speed learning algorithm for the quaternion neural network, substantially reducing network training time, while providing equivalent performance to quaternion backpropagation.