Ensemble of Deep Learning Models to Select Ovarian Cancer Patients for Bevacizumab Monoclonal Therapy

Bevacizumab monoclonal therapy, in combination with chemotherapy, is an FDA-approved treatment for ovarian cancer (OC), particularly for patients with epithelial ovarian cancer (EOC) patients. However, numerous studies have reported adverse effects associated with this treatment, including hypertension, bleeding, cardiac toxicity, and kidney issues in 70% of cases. Furthermore, therapy is costly, with a single cycle costing approximately $3,266. Due to these factors, pathologists carefully select patients for this therapy, often based on manual evaluation of Hematoxylin and eosin-stained (H&E) histopathology specimens under a microscope. This process is time-consuming, labor-intensive, and dependent on the availability of expert pathologists. Artificial intelligence (AI) and image-guided automated methods have been explored to address these challenges and improve patient care; however, current approaches only achieve a precision of 88. 2%. In this study, we propose an ensemble of deep learning models to predict the response of OC patients to Bevacizumab therapy based on their H&E-stained histopathology images to select OC patients for giving this therapy. Our proposed method achieved 97. 5% accuracy using 20X magnification images, significantly outperforming the existing method. Furthermore, this study comprehensively compared convolutional neural networks (CNN), transformers, and their ensemble models for this prediction.