Growth Plate Area Segmentation on Knee MRIs

The application of artificial intelligence plays a crucial role in modern medicine, helping to improve diagnostic process and patient care. This paper aims to explore the segmentation process on magnetic resonance imaging (MRI) of knee joint to segment the age-related parts of the knee joint, which are not sufficiently studied in age assessment research of young individuals. This paper presents and analyzes the results of MRI segmentation of the knee obtained using U-Net and U-net++ neural networks. Both networks were trained from scratch on a private dataset consisting of 530 2D knee MRI images. U-Net achieves a mean dice similarity coefficient value of 0.9415 and a mean Jaccardi index value of 0.8949 for knee bone segmentation, compared to U-Net++ with a mean dice similarity coefficient of 0.9280 and a mean Jaccardi index value of 0.8747.