Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Explainable and Conformal AI Model to Detect Temporomandibular Joint Involvement in Children Suffering from Juvenile Idiopathic Arthritis

Lena Todnem Bach Christensen, Dikte Straadt, Stratos Vassis, Christian Marius Lillelund, Peter Bangsgaard Stoustrup, Ruben Pauwels, Thomas Klit Pedersen, Christian Fischer Pedersen

TL;DR

The paper addresses the challenge of early TMJ involvement detection in children with Juvenile Idiopathic Arthritis by developing an explainable AI workflow based on Random Forest classification of longitudinal clinical exams. It integrates uncertainty estimation via MAPIE conformal prediction and feature-wise explanations with SHAP, while exploring temporal structures through IID, temporal segmentation, and lagged-feature strategies. Key findings show that, when using temporal segmentation, the model can identify TMJ involvement within two years of the first visit with precision 0.86 and sensitivity 0.70, and incorporating lag features modestly improves sensitivity. The approach demonstrates potential as a clinically useful decision-support tool to enable earlier, personalized TMJ management in JIA, though external validation is recommended.

Abstract

Juvenile idiopathic arthritis (JIA) is the most common rheumatic disease during childhood and adolescence. The temporomandibular joints (TMJ) are among the most frequently affected joints in patients with JIA, and mandibular growth is especially vulnerable to arthritic changes of the TMJ in children. A clinical examination is the most cost-effective method to diagnose TMJ involvement, but clinicians find it difficult to interpret and inaccurate when used only on clinical examinations. This study implemented an explainable artificial intelligence (AI) model that can help clinicians assess TMJ involvement. The classification model was trained using Random Forest on 6154 clinical examinations of 1035 pediatric patients (67% female, 33% male) and evaluated on its ability to correctly classify TMJ involvement or not on a separate test set. Most notably, the results show that the model can classify patients within two years of their first examination as having TMJ involvement with a precision of 0.86 and a sensitivity of 0.7. The results show promise for an AI model in the assessment of TMJ involvement in children and as a decision support tool.

An Explainable and Conformal AI Model to Detect Temporomandibular Joint Involvement in Children Suffering from Juvenile Idiopathic Arthritis

TL;DR

The paper addresses the challenge of early TMJ involvement detection in children with Juvenile Idiopathic Arthritis by developing an explainable AI workflow based on Random Forest classification of longitudinal clinical exams. It integrates uncertainty estimation via MAPIE conformal prediction and feature-wise explanations with SHAP, while exploring temporal structures through IID, temporal segmentation, and lagged-feature strategies. Key findings show that, when using temporal segmentation, the model can identify TMJ involvement within two years of the first visit with precision 0.86 and sensitivity 0.70, and incorporating lag features modestly improves sensitivity. The approach demonstrates potential as a clinically useful decision-support tool to enable earlier, personalized TMJ management in JIA, though external validation is recommended.

Abstract

Juvenile idiopathic arthritis (JIA) is the most common rheumatic disease during childhood and adolescence. The temporomandibular joints (TMJ) are among the most frequently affected joints in patients with JIA, and mandibular growth is especially vulnerable to arthritic changes of the TMJ in children. A clinical examination is the most cost-effective method to diagnose TMJ involvement, but clinicians find it difficult to interpret and inaccurate when used only on clinical examinations. This study implemented an explainable artificial intelligence (AI) model that can help clinicians assess TMJ involvement. The classification model was trained using Random Forest on 6154 clinical examinations of 1035 pediatric patients (67% female, 33% male) and evaluated on its ability to correctly classify TMJ involvement or not on a separate test set. Most notably, the results show that the model can classify patients within two years of their first examination as having TMJ involvement with a precision of 0.86 and a sensitivity of 0.7. The results show promise for an AI model in the assessment of TMJ involvement in children and as a decision support tool.
Paper Structure (13 sections, 2 equations, 2 figures, 2 tables)

This paper contains 13 sections, 2 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Clinical data
  4. Preprocessing and feature engineering
  5. Classification analysis
  6. Uncertainty estimation
  7. Explainability
  8. Experiments and results
  9. Strategy 1: IID Data
  10. Strategy 2: Temporal segmentation
  11. Strategy 3: Lagged Features
  12. Discussion
  13. Conclusion

Figures (2)

  • Figure 1: Each examination of a JIA patient includes a screening, where the patient's symptoms, drug usage and other clinical findings are recorded. The dataset contains historical examinations of 1035 JIA patients spanning a 25-year period with 6154 longitudinal records in total.
  • Figure 2: SHAP summary plots. The $y$-axis displays features, while the $x$-axis represents SHAP values. Features are ranked by importance, determined by the mean of their absolute SHAP values, with higher positions indicating greater significance. Each feature is represented by individual points, with a point corresponding to a single examination. The position on the $x$-axis signifies the impact of that feature on the model's output. The color of each point indicates whether the patient had a low (blue) or high (red) value of that specific feature. All SHAP values are computed on the test set.