Deep learning based tumor detection and segmentation for automated 3D breast ultrasound imaging

Breast cancer is one of the widely diagnosed cancer in the world. However, the detection and segmentation of the tumor is a problem which still needs to be solved. Here we proposed U-Net and YOLO for the segmentation and detection for breast tumor detection in ABUS images. The algorithms were used for 2D images and got a dice score of 0.567 for segmentation and a mAP score of 0.554 for detection of tumor for the split of training data. For validation dataset, the dice score was 0.5388 for segmentation and a detection score of 0.3988 for the detection of tumor in ABUS images.