Improved Cervical Cell Segmentation Algorithm Based on Cascade Mask RCNN

Cervical cell early detection can timely discover abnormal cells and take treatment to avoid cervical cancer. With the development of deep learning technology, cervical cell smear detection technology has made rapid progress. However, current deep learning segmentation algorithms still have problems with inaccurate segmentation and low segmentation accuracy when dealing with high overlap of cervical cells. Therefore, we propose the cell overlap degree to measure the degree of cell overlap, in order to compare the segmentation algorithm more targetedly. At the same time, to address the misclassification problem of the traditional Cascade Mask RCNN algorithm, we propose an improved algorithm based on Cascade Mask RCNN. This algorithm uses a more advanced backbone network to improve the segmentation accuracy of the algorithm, introduces a cross-scale feature fusion structure to improve the network’s ability to extract cell cytoplasm edge feature information, and improves the ROI pooling method to further enhance the algorithm’s feature extraction ability. The improved algorithm is compared in experiments on the dataset, and the results show that our algorithm has certain effectiveness and progressiveness.