Tooth segmentation on x rays limited by image resolution not model type
Detection is solved, delineation is not: what governs tooth segmentation on panoramic radiographs
Computer Vision and Pattern RecognitionMachine Learning
Summary
Automatically identifying and outlining teeth in panoramic dental x rays is important for computer-assisted dental care. The authors collected a large, carefully annotated dataset of dental images to study what factors affect performance. They found that higher image resolution improves how precisely the tooth boundaries are drawn, while the choice of AI architecture has little effect. Efforts to improve segmentation accuracy should focus on obtaining clearer images rather than on changing model designs.
What this means in practice
- •For dental imaging software teams: Develop dental x ray analysis tools prioritizing higher resolution input images to improve tooth boundary delineation accuracy.
- •For medical device manufacturers: Design panoramic radiograph capture devices that maximize resolution to enable more precise automatic tooth segmentation.
Authors
Muhammad Rehan, Moaz Amjad, Syed Danial Ahmed, Mariam Adnan, Haider Ali
Abstract
Automatic tooth segmentation and FDI numbering on panoramic radiographs underpins computer-assisted dental diagnosis, yet which factors govern performance remains unclear. We assemble a corpus of 1,422 panoramic radiographs containing 42,142 expert-delineated tooth polygons across the 32-class FDI taxonomy, annotated by 30 dental practitioners and independently reviewed by two others, and use it to isolate input resolution, architecture and anatomical priors under a single evaluation protocol. First, resolution dominates: across a controlled 640/1024/1280 ablation, mask mAP50-95 rises 0.656 -> 0.710 -> 0.717 while mAP50 stays flat at ~0.982. Both gains are significant under a paired bootstrap over images (p < 0.001, p = 0.024); neither mAP50 change is distinguishable from zero. Added resolution buys boundary precision, not detection. Second, architecture is nearly irrelevant in-domain: a query-based transformer with 2.1x the parameters is statistically equivalent to a one-stage detector (95% CI [-0.0064, +0.0064]), only marginally better under domain shift, 5.5x slower on CPU and not executable under standard ONNX runtimes. Third, three targeted interventions fail: a LoRA-adapted self-supervised encoder underperforms, a promptable foundation segmenter degrades masks by 39%, and globally optimal anatomical label assignment yields +0.0007 despite correcting a constraint violated in 40% of out-of-domain predictions. Zero-shot transfer to an independent multi-centre cohort, verified overlap-free, costs 62% of mask mAP50-95 but only 18% of mAP50, reproducing the dissociation. Decomposing masks along the tooth axis localises the residual error to the apical third. Boundary precision is therefore the binding constraint, and effort is better directed at resolution and acquisition diversity than at architectural novelty.