Embedding Clinical Diagnostic Logic in Deep Learning for Photographic Index of Orthodontic Treatment Need Classification

Background and aims Index of Orthodontic Treatment Need (IOTN) is a popular screening tool comprising an Aesthetic Component (AC) and a Dental Health Component (DHC), derived from a hierarchical assessment of five malocclusion traits (MOCDO: Missing teeth, Overjet, Crossbite, Displacement, and Overbite). Its photographic adaptation, photographic IOTN, enables practical remote screening. However, translating this multilevel clinical logic into deep learning (DL) remains challenging: whether models should predict DHC directly or use a composited, trait-based decision after assessing individual MOCDO components. Methods Retrospectively collected intraoral photographic sets (2020 sets, 5 views each) were annotated and segmented using SegmentAnyTooth. We compared two ConvNeXt-based architectures: a single-view direct baseline and a specific-view, trait-based approach utilizing four two-head models to predict MOCDO traits and AC. Treatment need was evaluated across primary (AC ≥ 6 or DHC ≥ 4), UK-NHS, and broad screening. Evaluation was performed on an independent test set ( n = 204), using paired bootstrap resampling (Δ, P values) and EigenGrad-CAM for interpretability. Results Both architectures achieved comparable AC classification. However, the trait-based approach outperformed the direct baseline for composite DHC assessment (macro F1: 0.77 vs 0.52, ΔF1 = +0.24, P < .001). For treatment need, the trait-based model demonstrated balanced performance (sensitivity: 0.91, precision: 0.92), whereas the direct baseline showed severe overprediction (precision: 0.78, Δprecision = +0.14, P < .001). This precision advantage was consistent across both NHS and broad screening thresholds ( P < .001). Conclusions Embedding clinical diagnostic logic into DL improves the accuracy, explainability, and precision of automated photographic IOTN triage across multiple thresholds. Clinical relevance Embedding the hierarchical MOCDO trait logic into DL models outperforms direct DHC prediction, preventing treatment severity overestimation and providing a more reliable tool for remote orthodontic screening.

Authors

Institutions

Publication Details

Journal
International Dental Journal
Published
2026-11-19
DOI
https://doi.org/10.1016/j.identj.2026.111356
Primary Topic
Orthodontics and Dentofacial Orthopedics
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Embedding Clinical Diagnostic Logic in Deep Learning for Photographic Index of Orthodontic Treatment Need Classification

Duy Le Nguyen, Guang Hong, Ding‐Han Wang, Ming‐Lun Hsu et al.
International Dental Journal
Orthodontics and Dentofacial Orthopedics
article

Embedding Clinical Diagnostic Logic in Deep Learning for Photographic Index of Orthodontic Treatment Need Classification

Duy Le Nguyen, Guang Hong, Ding‐Han Wang, Ming‐Lun Hsu, Hong Thi-Phuong Doan, Khoa Dang Nguyen, Hung Trong Hoang
article en

Abstract

Background and aims Index of Orthodontic Treatment Need (IOTN) is a popular screening tool comprising an Aesthetic Component (AC) and a Dental Health Component (DHC), derived from a hierarchical assessment of five malocclusion traits (MOCDO: Missing teeth, Overjet, Crossbite, Displacement, and Overbite). Its photographic adaptation, photographic IOTN, enables practical remote screening. However, translating this multilevel clinical logic into deep learning (DL) remains challenging: whether models should predict DHC directly or use a composited, trait-based decision after assessing individual MOCDO components. Methods Retrospectively collected intraoral photographic sets (2020 sets, 5 views each) were annotated and segmented using SegmentAnyTooth. We compared two ConvNeXt-based architectures: a single-view direct baseline and a specific-view, trait-based approach utilizing four two-head models to predict MOCDO traits and AC. Treatment need was evaluated across primary (AC ≥ 6 or DHC ≥ 4), UK-NHS, and broad screening. Evaluation was performed on an independent test set ( n = 204), using paired bootstrap resampling (Δ, P values) and EigenGrad-CAM for interpretability. Results Both architectures achieved comparable AC classification. However, the trait-based approach outperformed the direct baseline for composite DHC assessment (macro F1: 0.77 vs 0.52, ΔF1 = +0.24, P < .001). For treatment need, the trait-based model demonstrated balanced performance (sensitivity: 0.91, precision: 0.92), whereas the direct baseline showed severe overprediction (precision: 0.78, Δprecision = +0.14, P < .001). This precision advantage was consistent across both NHS and broad screening thresholds ( P < .001). Conclusions Embedding clinical diagnostic logic into DL improves the accuracy, explainability, and precision of automated photographic IOTN triage across multiple thresholds. Clinical relevance Embedding the hierarchical MOCDO trait logic into DL models outperforms direct DHC prediction, preventing treatment severity overestimation and providing a more reliable tool for remote orthodontic screening.

International Dental JournalVol. 77(1)
National Yang Ming Chiao Tung University (TW), Pham Ngoc Thach University of Medicine (VN), Tohoku University (JP), University of Medicine and Pharmacy at Ho Chi Minh City (VN), Liaison Center for Innovative Dentistry, Graduate School of Dentistry, Tohoku University (JP)
Orthodontics and Dentofacial Orthopedics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon