Applying AI to Digital Pathology for Cancer Research & Clinical Development

Dr. Ralf Huss, Chief Medical Officer at Definiens, explains how the innovative combination of artificial intelligence (AI) and expert knowledge is critical for driving the successful adoption of digital solutions in cancer research. 

In an age of increasingly sophisticated technologies for data collection and analysis, have we achieved better insights? 

Over the last decades, we’ve witnessed incredible developments that have expanded our understanding of cancer research. And yet, even as sophisticated technologies like artificial intelligence have gone from single cell analysis to multidimensional disease modelling, it’s clear now that we need to evolve in how we identify meaningful insights and separate good knowledge from distracting information. That’s why AI, much like two sides of a coin, provides both a cognitive-based approach as well as hypothesis-free intelligence that when used appropriately, gives us a better understanding of cancer.

How can AI help achieve the ambitious goal of linking existing and proven expert knowledge to novel, hypothesis-free insights?  

Fortunately, AI not only provides single solutions but can also be flexibly applied depending on the problem. But what does it take to technically and clinically validate hypotheses generated from big data sets including tissue images and genomics, obtained from experiments and clinical trials? Indeed, on one side, Definiens’ cognition solutions naturally integrate machine learning and random forest decision trees with supervised guidance by experts. On the other side, semi- or unsupervised solutions like deep learning or convolutional neural networks add another layer of insights. Even then, it still requires the expertise of pathologists, data specialists, and cancer researchers for plausibility checks, and to suggest robust validation paths into the clinical practice.

AI-driven image analysis tools have fueled the advancement of cancer research - but can AI also be used for clinical development?

Cancer research depends heavily on the understanding of the tumor phenomenology and its implicit heterogeneity. As such, unlocking the cancer phenome should be a prerequisite for treatment decisions, particularly in the field of cancer immunotherapies. But it requires the standardization of data acquisition through diagnostic or panel tests; these allow the comparison and integration of image and other -omics data on a routine basis within the clinical setting. The spatial relationship of cancer and immune cells in their tumor microenvironment together with biologically-relevant information, active pathways, and the help of AI-driven analysis, will drive the intelligent matching of patients with the most appropriate therapies.

AI assists the discovery and development of new cancer treatments - but how does AI depend on humans to be effective?

Put simply, AI helps physicians prescribe effective therapies for the maximum benefit of the patients. Thus, AI makes good doctors even better doctors. And yet, for AI to be successful, it’s dependent on the availability of well curated data sets, validated insights, and generated hypotheses by experts in well-designed studies or clinical trials. It is this human scrutiny and professional intervention that expedites the implementation of AI into the advancement of biological and medical science.

www.global-engage.comwp-contentuploads201811154356664949977425-2-1

Dr. Ralf Huss 

Dr. Huss has more than 20 years of training and experience in histopathology and cancer research, and served previously as Global Head of Histopathology and Tissue Biomarker at Roche Diagnostics, where he was involved in identifying new tissue biomarkers to stratify cancer patients as part of a global Personalized Healthcare strategy. Professor Huss also co-founded the biotech company APCETH for which he remains a member of the management team. In addition, Professor Huss holds national and international academic appointments at the Ludwig-Maximilians-University Munich (Germany) and the Wake Forest Institute for Regenerative Medicine (USA).

 

Interested to learn more about AI and biomarker analysis using Tissue Phenomics? Download a free copy of the first chapter of our book here. 

This article originally appeared in Cancer Research Industry News magazine distributed at AACR 2019.