Deep Learning Predicts Chemotherapy Response in Ovarian Cancer from CT Scans

Researchers have created an innovative deep learning method to predict how patients with ovarian cancer will respond to neoadjuvant chemotherapy, using pre-treatment contrast-enhanced CT scans. Ovarian cancer is the most lethal gynecological cancer, with about 60% of patients diagnosed at advanced stages and a 5-year survival rate around 30%. Early identification of those who won’t respond can help steer clear of ineffective treatments and avoid delays in surgery. The method utilizes 3D lesion masks and a specialized image encoder to create detailed volumetric embeddings. It combines classification loss with contrastive regularization and hard-negative mining to better distinguish between responders and non-responders. While the method was based on a retrospective dataset, specific details about the cohort are not mentioned. This approach addresses a crucial gap in ovarian cancer treatment.