Oncology Decision Support in Ovarian Cancer: Artificial Intelligence-Based Pathomics to Identify Platinum-Resistant Epithelial Ovarian Cancer

Poster Number: P189
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Deep Learning, Machine Learning, Clinical Decision Support, Precision Medicine
Primary Track: Applications
Programmatic Theme: Clinical Informatics

People with ovarian cancer typically present at an advanced stage, and the standard of care is extensive surgery followed by platinum-based chemotherapy. However, approximately 15% of people will be platinum-resistant, meaning that they will relapse within six months of their surgery. Such patients require different, more aggressive chemotherapy than the standard, which could increase their time to progression and survival rate. To this end, we propose a multimodal deep learning framework to identify people with platinum-resistant epithelial ovarian cancer and resistance to bevacizumab, a recently FDA approved targeted molecular therapy for ovarian cancer. In particular, we take into account not only gigapixel histopathology images, but also patient clinical variables under a multi-modal learning framework. Evaluation is performed on two ovarian cancer benchmarks: a public The Cancer Imaging Archive Ovarian Bevacizumab Response (TCIA-OBR) dataset and a in-house Predictovar dataset, a retrospective cohort of patients from the Karolinska University Hospital in Stockholm, Sweden. We achieve high prediction test accuracies on treatment resistance of 0.83 and 0.78 on TCIA-OBR and Predictovar dataset, respectively. We further delve into the interpretability of the proposed model by visualizing the feature interactions from patient clinical variables such as patient age and patch-level contributions to the model's predictions. Our proposed approach is a step toward understanding the factors that influence treatment effectiveness in ovarian cancer patients.

Speaker(s):
Emily Nguyen, PhD student
University of Southern California