Fig. 1
Overview of the ovarian cancer treatment outcome prediction system. (A) The overall workflow of our framework to predict treatment outcomes in the diagnostic stage. (B) Diagram showing how we select thresholds to determine the short PFS (< 8 months) and long PFS groups (> 12 months) to train and evaluate our model. We also considered the typical threshold for short PFS in practice (< 6 months) and performed zero-shot and few-shot evaluations. (C) Contrastive pre-training framework with view contrast and location contrast to pre-train a ResNet-50 backbone. (D) Data augmentation strategy to form a bag of augmented image groups for each patient and the outcome prediction model using a transformer encoder to train on sequences of image features.
