This study focuses on the diagnostic techniques of temporomandibular joint disorder (TMD) and proposes a fully automatic joint disc and condyle detection and segmentation system based on deep learning. The aim is to enhance the diagnostic accuracy of TMD through magnetic resonance imaging (mri). At the same time, develop a practical tool software for viewing dicom medical images. A total of 25 groups of MRI images were collected in this study, and each group of images included 8–10 MRI images. In this study, an improved algorithm based on U-Net was adopted for the segmentation of the articular disc, and it was compared and analyzed with the traditional U-Net algorithm. The results show that the algorithm used in this study performs excellently in the Dice coefficient and is superior to the traditional U-Net algorithm. This indicates that this algorithm has higher accuracy when dealing with the task of joint disc segmentation. The research results show that the segmentation method based on machine learning is expected to be used in clinical practice for the assessment of TMD, providing support for the diagnosis and treatment of related diseases.