Deep Learning for TMJ Disc Segmentation on MRI

Deep Learning for TMJ Disc Segmentation on MRI

Authors:

Shota Ito; Yuichi Mine; Yuki Yoshimi; Saori Takeda; Akari Tanaka; Azusa Onishi; Tzu‑Yu Peng; Takashi Nakamoto; Toshikazu Nagasaki; Naoya Kakimoto; Takeshi Murayama; Kotaro Tanimoto

Summary:

Study evaluating three deep learning–based semantic segmentation algorithms—3DiscNet, U‑Net, and SegNet‑Basic—for automatic segmentation of the temporomandibular joint (TMJ) articular disc on MRI. Using 217 sagittal MR images (normal discs + displaced discs), 3DiscNet and SegNet‑Basic showed superior performance compared to U‑Net, achieving higher Dice, sensitivity and PPV. The results provide proof‑of‑concept that deep learning can support clinical assessment of TMD.

Check the original publication:

Key Words:

TMJ; MRI; Articular Disc; Segmentation; Deep Learning; 3DiscNet; U‑Net; SegNet‑Basic

Extracted Data:

Year:

2022

Modality:

MRI (Temporomandibular Joint)

Dataset:

217 sagittal proton‑density MRI slices (10 patients + 10 controls)

Dataset Split:

Training 80% / Test 20% for each subgroup (normal, displaced, combined)

Network Architecture:

3DiscNet (encoder‑decoder); U‑Net; SegNet‑Basic

Metrics:

Dice, Sensitivity, PPV (Best: Dice 0.74 for SegNet‑Basic)

AP - Professional Qty:

3 experts (2 orthodontists + 1 oral and maxillofacial radiologist)

AP - Supervisor Presence:

Yes

AP - Experience Level:

6–25 years

AP - Expertise Area:

Orthodontics; Oral and Maxillofacial Radiology

AP - Tool or System:

ImageJ for segmentation; Python/Keras/TensorFlow for training

ML Task:

Semantic Segmentation (TMJ Articular Disc)

Project Objective:

Develop a fully automated TMJ disc segmentation system using deep learning.

Clinical Relevance:

Provides automated and reproducible segmentation of the TMJ disc on MRI, supporting diagnosis of temporomandibular disorders (TMD) and reducing variability in clinical assessment.

Watch full video on YouTube