Optimizing Operating Points for High Performance Lesion Detection and Segmentation Using Lesion Size Reweighting

There are many clinical contexts which require accurate detection and\nsegmentation of all focal pathologies (e.g. lesions, tumours) in patient\nimages. In cases where there are a mix of small and large lesions, standard\nbinary cross entropy loss will result in better segmentation of large lesions\nat the expense of missing small ones. Adjusting the operating point to\naccurately detect all lesions generally leads to oversegmentation of large\nlesions. In this work, we propose a novel reweighing strategy to eliminate this\nperformance gap, increasing small pathology detection performance while\nmaintaining segmentation accuracy. We show that our reweighing strategy vastly\noutperforms competing strategies based on experiments on a large scale,\nmulti-scanner, multi-center dataset of Multiple Sclerosis patient images.\n