The proposed framework leverages weakly supervised multiple-instance learning (MIL) to reduce false-positive predictions. A novel morphology-based slide aggregation method is introduced to improve ...