Region‐based two‐stage MRI bone tissue segmentation of the knee joint

In medical image segmentation, the neural network structure of U-Net family has demonstrated sufficient advantages. However, MRI images have different scan parameters and times, resulting in feature representation images. Furthermore, there is a great class imbalance between bone cartilage tissues knee To address these issues, region-based two-stage tissue segmentation proposed this paper. The makes full use location characteristics three types joint uses architecture with modified U2-Net backbone to segment tissue. divided into two phases, first phase simple coded decoding for saliency detection obtain positional regional relationships tissues, second consisting 2 U2-Net, one segmenting patella associated other femur, tibia cartilage. algorithm was tested variety data verify effectiveness algorithm.