Field of Expertise
Computational Pathology
Expertise
Expertise: We are applying to this challenge as employees of the Institute of Cancer Research (ICR), a world-leading research institute with track record of working with industry. The lead applicant, Dr Yinyin Yuan and three consultant pathologists are exploring all possible avenues to find the best available route for our AI-based technology to reach the market for patient benefit. We are open to working with industry through a licensing co-development route, as well as other commercialisation pathways (e.g. venture formation), should right opportunities arise. We also have plans to form an ICR-backed AI-histology spinoff, to be incorporated soon. Team: -Dr Khalid AbdulJabbar, PhD, Postdoctoral Fellow in Computational Pathology at ICR. -Dr Yinyin Yuan, PhD, Team Leader and Reader in Computational Pathology at ICR, London. Dr Yuan has over 10 years of experience in digital pathology and over 15 years in machine learning. She is a well-known pioneer in the field of spatial histology and an expert on computational genotype-phenotype integration in cancer. Advisors and opinion leaders: -Prof. John Le Quesne, MD PhD, Beatson Institute. John has been the lead pathologist for TRACERx, the largest single study yet funded by CRUK. He is an expert in the diagnosis of lung cancer, and in the development of novel histological assays and image analysis methods to improve cancer prognosis and diagnosis: predictive biomarkers and mRNA. -Dr Roberto Salgado, MD PhD, Peter MacCallum Cancer Centre & GZA-ZNA Hospitals, Antwerp. Roberto is a certified pathologist for over 15 years. He co-chairs the TILs immune-oncology group and expert of ML-pathology guidelines. -Dr Hugo Horlings, MD PhD, Netherlands Cancer Institute. Hugo co-founded ‘SlideScore’, a user-friendly scoring interface for pathologists, he brings in hands-on knowledge on how to sell our technology to practicing pathologists. He has strong skills in both cancer molecular biology and genomics. We have been close collaborators for years; we have developed various AI tools for different problems targeting over 10 different cancer types. Our advisors are key opinion leaders and they bring in strategic access to thousands of samples, several high-quality international trials data; a very strong edge over competitors, which together with our AI technology and Pfizer-IBM’s industry expertise could be further leveraged for patient benefit. Problem and solution: The real problem that still needs to be addressed lies around the translation of AI for ‘accurate diagnostics and treatment selection’ for routine clinical use. We need a solution that any clinician world-wide can use and interpret, not only to experts who are involved in digital pathology. Cancer patients need universally optimal care; computational pathology, at the moment, delivers too little in the clinic to achieve this goal. Basic biological laws tell us this is bound to change. Even non-experts could predict that by closely studying the tumor tissue, marking every cell and its neighbour's role in suppressing predatory cancer cells, seeing the big picture while examining the micro-picture at single-cell level, will only lead to better understating of disease and drug – the question is how? Current computational pathology applications are too coarse, they often rely on mixing multi-modality datasets into one big basket (AI engine), to make a prediction on a histological image. Biological context is often ignored. Subsequently, these approaches will rarely generalize to a new set of patients, and almost never make their way to the clinic and make a real difference to the way we treat cancer. Moreover, the most important biological trait of this disease is ignored: cancers evolve. Cancer is a genetic disease, it evolves, it adapts, and it escapes immune surveillance. This unstoppable force of nature is why we can’t generally treat cancer once it is spread. To defeat cancer, we need to outsmart it. One way is to deploy AI to “siege” the tumor by scrutinising its environment where it is growing; deploy AI not merely to recognise and score the tumor, but also understand its influences. Our proprietary technology stems from developing AI that targets cancer evolution directly from routine histology slides – no other lab in the world has connected these two disciplines in the way we have done it. In a recent study, we found that the geospatial immune landscape is directly complicit in shaping cancer evolution, and it can be measured. This integration puts us in a unique position: AI application for drug discovery using a Darwinian strategy; looking closely at the ecological-evolutionary context while treating cancer. On the delivery front, clinical translation comes with its own unique challenges. As a team, we have the right set of skills: AI and software know-how, large-scale slide digitization and data management, spatial histology, and pathology experience, which is further strengthened and complemented by our clinician advisors. Together with colleagues from the Royal Marsden Hospital, our lab is in the final stages of drafting plans for an end-to-end AI-histology deployment in the clinical setting. We aim to test our AI technology to automate immune scoring for triple-negative breast cancers (TNBC), an aggressive breast cancer subtype with a considerable lack of targeted treatments and biomarkers. In addition, we will test the solution in 7 external, phase 3 clinical trials of TNBC, in close partnership with International TIL Working Group, to commence Q4 2020. We aim to provide a pioneering example of AI and digital pathology deployment integrated with clinical research in the Royal Marsden. This will provide additional evidence of clinical utility of AI-based pathology, and ultimately, further develop our AI as a novel biomarker. As computational biologists with a real passion for clinical impact, we are looking to collaborate with commercial partners with expertise in digital health, therapeutic development and regulations (both medical devices and pharmaceutical), who could help us further develop our AI technology with a view to commercialising and introducing it in healthcare settings for patient benefit. In our view, IBM and Pfizer represent important partners who are ideally positioned to help us achieve this goal from the perspective of digital health and biomarker/therapeutic development, respectively. Given our hands-on experience with gastrointestinal samples from the unique Seattle cohort (see vetting form), and our international efforts to deploy AI for clinical use, we are well positioned to participate in this innovation challenge to further fine-tune our AI-based technology. We are open-minded and are willing to work with Pfizer/IBM and NHS to make further advancements in technical development and commercialization around the world, starting in the UK. We believe computational pathology powered by impactful AI will ultimately be used as a biomarker to guide therapeutic decision making, such as de-escalation of patients unsuitable for a neo-adjuvant treatment, thereby reducing toxicity and costs.