Quality checker improves trust in dental X-ray segmentation results

Beyond Ground Truth: A Heuristic Unsupervised Quality Assurance Engine for Diagnosing Calibrational Blindness in Dental Segmentation

Computer Vision and Pattern Recognition

Summary

Dental X-ray images need to be segmented accurately to help dentists make diagnoses, but deep learning models often give wrong segmentations with high confidence, which is misleading. The authors created a tool called HUS-QA that checks the quality of the segmentation without needing a correct answer to compare against. This tool looks at features like confidence, edge sharpness, and shape connectivity to guess how good the segmentation is. They found that this method correlates well with actual segmentation quality and works better than just trusting the model's confidence score. This can help doctors trust automated diagnoses more in real clinical settings.

What this means in practice

  • For clinical imaging teams: Automatically flag unreliable dental X-ray segmentations before clinical review using a no-ground-truth quality check.
  • For medical device developers: Integrate HUS-QA into dental imaging software to improve reliability warnings when ground-truth masks are unavailable.$Commercial implications: Enables product features that alert users to unreliable dental segmentations, enhancing clinical confidence and safety.

Tested on one dataset.

Authors

Fatematuj Johura Mim

Abstract

In real-world clinical settings, dental X-ray segmentation using deep learning methods suffers from reliability issues. Even when the segmentation quality is bad, deep learning models show high confidence in their predictions (calibrational blindness). Moreover, in real use, doctors do not have ground-truth masks to check their accuracy. To address this problem, we introduce a simple quality-checking engine called HUS-QA with three stages. We train it on the PRAD (Periapical Dental X-ray) dataset to build our SimCLR self-supervised ResNet-18 U-Net model, which generates the segmentation mask and then evaluates its quality. Our HUS-QA checks the segmentation mask quality without a ground-truth mask. Prediction confidence, edge sharpness, shape connectivity, and area-size penalty are the simple indicators considered in this research. Results show that standard confidence methods fail (r = -0.1394, p = 1.7864e-3, ROC AUC = 0.4275). By contrast, our HUS-QA shows a positive correlation with actual Dice scores (r = 0.4647, p = 3.7510e-28) and achieves a strong ROC-AUC of 0.8405. Therefore, HUS-QA serves as a practical tool for clinical quality control.
View PDFOpen arXiv