Abstract
Automated assessment of temporomandibular joint osteoar- thritis (TMJ OA) from cone-beam computed tomography (CBCT) re- mains limited by reliance on manually defined regions of interest and by uncertainty about how predicted anatomy should be encoded for 3D classification. This paper presents AutoTMJ-OA, a multi-stage CBCT framework for detecting radiographic osteoarthritic features as present or absent. Internal full-volume CBCT scans undergo sagittal TMJ local- ization, 3D ROI extraction, and SegResNet-based mandible segmenta- tion; the public University of Michigan Deep Blue cohort enters directly at the ROI stage. The combined cohort included 216 patients and 432 joint-level samples, split at the patient level into 252 training, 87 valida- tion, and 93 test joints. Four inputs generated from the same ROI and mandible mask—volume only, volume plus mask, masked volume, and masked volume plus mask—were evaluated at five cubic input sizes from 963 to 2563 voxels using an identical classifier architecture and train- ing protocol. Masked CBCT volume achieved the highest test-partition AUC at four of five input sizes, with the highest observed performance at 1923 voxels (AUC 0.938; F1 0.926). Adding an explicit mask chan- nel to an already masked volume did not improve performance. Because all configurations were evaluated on the same test partition, results are interpreted descriptively rather than as a pre-specified confirmatory com- parison. These findings suggest that anatomical background suppression may contribute more to TMJ OA feature detection than an additional explicit shape channel.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ODIN_021.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=Cb5nTl8Awu
BibTex
@InProceedings{TriShr_From_MICCAISAT2026,
author = { Trivedi, Shradhdha AND Sojitra, Vrundan AND Padilla, Mariela},
title = { { From TMJ Localization to Detection of Osteoarthritic Features: An Anatomically Guided 3D Deep Learning Pipeline for Cone-Beam CT } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17270},
month = {pending},
page = {pending}
}