Fig. 1: Heatmap visualisation of image-based deep learning predictions of malignancy.

Visual explanations of DL models are definitely important for qualitative review and clinical relevance, namely irregular solid components, projections, and areas with abundant blood flow signals. a Carcinosarcoma of a 44-year old female; b high-grade serous carcinoma of a 65-year old female; and (c) hydrosalpinx of a 49-year old female that was misdiagnosed by all readers but showed a low probability of malignancy in the heatmap. In the first row of each case, the first two images are B-mode images, and the following one is the colour Doppler image. The images in the second row are their corresponding heatmaps.