DENTEX: Dental Enumeration and Tooth Pathosis Detection Benchmark for Panoramic X-ray

TL;DR

The paper introduces DENTEX, a MICCAI 2023 challenge to benchmark AI for dental enumeration and tooth pathology detection on panoramic X-rays. It presents a three-level hierarchical dataset and a two-phase evaluation that rewards learning from partial labels and robust multi-label prediction of quadrant, enumeration, and diagnosis. Through a thorough analysis of baseline, SOTA, and participating methods, the study shows segmentation-aware and diffusion-based architectures excel in enumeration and diagnosis, respectively, while ensembles offer robust practical performance. The findings provide architectural guidance for clinically usable AI tools in dentistry and highlight future directions such as larger diverse datasets and multi-rater validation.

Abstract

Panoramic X-rays are frequently used in dentistry for treatment planning, but their interpretation can be both time-consuming and prone to error. Artificial intelligence (AI) has the potential to aid in the analysis of these X-rays, thereby improving the accuracy of dental diagnoses and treatment plans. Nevertheless, designing automated algorithms for this purpose poses significant challenges, mainly due to the scarcity of annotated data and variations in anatomical structure. To address these issues, we organized the Dental Enumeration and Diagnosis on Panoramic X-rays Challenge (DENTEX) in association with the International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) in 2023. This challenge aims to promote the development of algorithms for multi-label detection of abnormal teeth, using three types of hierarchically annotated data: partially annotated quadrant data, partially annotated quadrant-enumeration data, and fully annotated quadrant-enumeration-diagnosis data, inclusive of four different diagnoses. In this paper, we present a comprehensive analysis of the methods and results from the challenge. Our findings reveal that top performers succeeded through diverse, specialized strategies, from segmentation-guided pipelines to highly-engineered single-stage detectors, using advanced Transformer and diffusion models. These strategies significantly outperformed traditional approaches, particularly for the challenging tasks of tooth enumeration and subtle disease classification. By dissecting the architectural choices that drove success, this paper provides key insights for future development of AI-powered tools that can offer more precise and efficient diagnosis and treatment planning in dentistry. The evaluation code and datasets can be accessed at https://github.com/ibrahimethemhamamci/DENTEX

Figures (7)

• Figure 1: The hierarchical organization of the annotated data used in the DENTEX. The data is structured into three levels: (a) quadrant-only for quadrant detection, (b) quadrant-enumeration for tooth detection, and (c) quadrant-enumeration-diagnosis for abnormal tooth detection.
• Figure 2: Desired output from the final model: Illustrating well-defined bounding boxes for abnormal teeth. The corresponding quadrant (Q: 1-4), enumeration (N: 1-8), and diagnosis (D: Caries, Deep caries, Periapical lesion, Impacted) labels are also displayed.
• Figure 3: Performance metrics of the methods on tasks of (a) Quadrant Detection, (b) Enumeration Detection, and (c) Diagnosis Detection.
• Figure 4: Performance Rank Across All Tasks and Final Mean Rank of Methods. In ascending order of final mean rank (lower the better). Each cell contains the rank (1-11) of a method for a specific metric. Darker colors indicate a better rank.
• Figure 5: Qualitative performance of submitted models on an unambiguous pathology class: Impacted third molars. This figure illustrates a common finding where models across the performance spectrum achieved success, from the top-ranked He L. to the lowest-ranked Dascalu T.