Mitosis Detection Using Image Segmentation and Object Detection

Abstract

The World Health Organisation(WHO) identifies that in women, the second most cancer deaths are caused by Breast cancer[1]. This paper presents various approaches for Mitosis detection on publicly available MITOS data set and DSB (Data Science Bowl). The process involves using a U-Net architecture consisting of convolution and deconvolution layers to perform the image segmentation. On the segmented image, YOLO algorithm is used to perform the object detection, thus forming bounding boxes around the nuclei. The next task involves the classification of nuclei into either mitotic or amitotic which is achieved with help of one class SVM. The results achieved on the data sets were able to prove that the process followed got good results for mitosis detection on histology images.