Semi-automated Image Analysis for Histology
PI: Helen Byrne
Department: Mathematical Institute
Modern imaging techniques generate detailed images of tissue biopsies, typically at the cell scale, each image containing hundreds of millions of pixels. Manual analysis of such images is time-consuming, and diagnoses (even by experienced pathologists) may not be consistent. As more pathology departments become fully digitalised, demand for high-throughput methods to process scanned histological (H&E) and immunohistochemistry (IHC) slides is increasing.
Our automated image analysis software could play a significant role in meeting this demand. By combining image compression techniques with machine learning, our software enables rapid extraction from large H&E and IHC images of quantitative data that are accurate, robust and reproducible. Efficient analyses of biopsies using our approach will assist with more rapid delivery of effective and personalised treatments to patients. This will improve quality of life for patients and reduce the economic burden on healthcare providers.
Before our software can be used in a clinical or industrial setting, we must systematically test its performance against existing commercial software and experienced pathologists and establish user confidence in its diagnostic power. The main aim of this project is to perform such a benchmarking study with pathologists and clinicians from Oxford and scientists from the pharmaceutical company Roche. Our project partners will also provide feedback on how to improve the software so that non-experts can easily use it and so that it keeps pace with imaging developments. We will investigate the feasibility of adapting the software to simultaneously analyse multiple types of images and generate multiscale datasets that contain patient-specific tissue, cell and subcellular information and/or to determine how new treatments for diseases such as cancer act. In the longer term, we aim to integrate these multiscale data with computational models that compare different treatments for individual patients.