Segmentation ability map: Interpret deep features for medical image segmentation

Sheng He, Yanfang Feng, Patricia Ellen Grant, Yangming Ou

Abstract: Highlights•We present a ProtoSeg method for interpreting the deep features learnt by the U-Net for medical image segmentation.•The proposed ProtoSeg can compute the segmentation ability map on each deep feature with a quantitatively measurement of the feature’s segmentation ability.•Our method can provide new insights for interpreting the explainable AI systems for medical image segmentation.