A Field in Transition
How close are we to the full-scale clinical deployment of AI? We round up some of the most exciting stories in computational pathology today
Luke Turner | | Quick Read
What’s the Holdup?
Although the vast majority of labs use digital viewing systems for quality assurance and professional development, the use of digital pathology in routine workflows remains relatively uncommon, and whole-scale adoption is even rarer. A key challenge hampering adoption? Cash-strapped providers are often reluctant to back digital pathology. “Linking the business case to lack of staffing helps, but the efficiency savings in existing senior staff time, as well as multidisciplinary team time savings are the real levelers,” says Jo Martin, President of the Royal College of Pathologists.
Predicting Prognosis
Digital pathology has opened a number of doors for the development of image analysis tools to detect disease. But what if we could use digital techniques to predict recurrence risk, disease aggressiveness, and long-term survival? Hoping to offer alternatives to molecular assays, Anant Madabhushi and his team have already made technological breakthroughs in several disease domains, including breast, prostate, and lung cancer. Their ultimate aim? Routine use of computer vision algorithms to identify and extract histological biomarkers visible in the tissue.
Putting the Algorithms in Pathology
Although numerous studies have demonstrated deep learning algorithms’ potential to match – or even better – the accuracy of highly trained diagnostic experts, they haven’t yet entered routine clinical use. “I get very envious when I see fields such as ophthalmology and radiology already using FDA-approved deep learning algorithms – we need to bring pathology to this level,” says Johan Lundin, Research Director at the Institute for Molecular Medicine. To drive development and broaden the potential applications of AI, Lundin calls for more accurate annotation, wider sharing of annotated images, and the use of supervised learning to develop algorithms for less subjective endpoints.
Centers of Excellence
A number of UK centers backed by government funding are striving to advance the clinical deployment of AI for disease diagnosis. David Harrison, Lead of the iCAIRD Consortium in Scotland, believes that the key to success is to use digital pathology to transform workflows and inform decisions, rather than to make a diagnosis. The NPIC Consortium in Leeds has similar ambitions, aiming to digitize 760,000 slides per year with the ultimate goal of full digitization of National Health Service (NHS) laboratories. Meanwhile, the PathLAKE Consortium is building a data lake of “real-world” NHS data that will culminate in the world’s largest repository of annotated pathology images.
Into the Deep for Histopathology
Jeroen van der Laak, Group Leader of the Computational Pathology Group at Radboud University Medical Center, has explored the potential of deep learning for pattern recognition in digitized breast cancer slides. “The most straightforward application is automated assessment of the lymph node status, which may support tumor staging,” says van der Laak. The team has also developed algorithms that automatically recognize and count mitotic figures, aiding breast cancer grading. They hope that, with the discovery of novel prognostic biomarkers, these algorithms will drive the advancement of precision medicine.
