CT Recognition for Liver and Lesions

This article uses a two-step method for liver and lesions recognition, and our main purpose is to identify liver lesions. Therefore, the first step does not need much resources. After the segmentation of the liver is performed, and the recognition of the lesions in the second step requires fine identification. Therefore, it is necessary to identify with U-net, which has a high recognition accuracy. This paper proposes a method that can effectively reduce the complexity of the algorithm and ensure the accuracy. The first step is to use the traditional segmentation algorithm level set to segment the liver, and the second step to increase the segmentation effect by increasing the depth of the U-net and reducing the size of the convolution kernels. 89.3% of the lesions segmentation accuracy can be obtained on commercial medical institutions’ CT, and the lesions segmentation accuracy on the open data set can reach more than 90%, which may meet the needs of the assistant doctors.