Automatic Detection And Visualization Of Temporomandibular Joint Effusion With Deep Neural Network

This study explores the use of deep learning-based automatic detection of TMJ effusion in patients with temporomandibular disorder using MRI images. Results show that the fine-tuning model with proton density images has the highest diagnostic performance, with excellent specificity but lower sensitivity compared to human experts. Grad-CAM visualizations also support the model’s focus on effusion in the TMJ area. However, the model’s performance did not improve when using T2-weighted images in addition to PD images.