Clinical Report: Integrative Machine Learning Approaches for Differentiating Pediatric MS
Overview
This study evaluates the efficacy of multimodal deep learning models using optical coherence tomography (OCT) to differentiate pediatric multiple sclerosis (POMS) from non-inflammatory neurological conditions. The early fusion model demonstrated superior diagnostic performance, achieving an AUC of 0.90 and an accuracy of 87%.
Background
Pediatric-onset multiple sclerosis (POMS) is a rare but critical condition that requires early diagnosis for effective intervention. The optic nerve is often the first site of inflammation in MS, making it a focal point for diagnostic imaging. Optical coherence tomography (OCT) has emerged as a valuable tool in assessing retinal changes associated with MS, yet its application in pediatric cases remains underexplored.
Data Highlights
Model
AUC
Weighted F1
Macro F1
Accuracy
Early Fusion
0.90
0.87
0.77
87%
Best Unimodal Feature-based (SVC)
0.84
0.85
0.73
85%
Best Image-based (ResNet101 with SVC)
0.79
0.84
0.70
87%
Late Fusion
N/A
N/A
N/A
82%
Key Findings
The early fusion model outperformed unimodal and late fusion models in diagnosing POMS.
OCT-derived features correlate with disease severity and visual dysfunction in MS.
POMS accounts for approximately 3–5% of all MS cases, highlighting the need for effective diagnostic tools.
Multimodal learning captures complementary patterns associated with MS pathology.
Diagnostic ambiguity in pediatric MS necessitates integrating multiple data sources for improved accuracy.
Clinical Implications
The findings suggest that integrating OCT with machine learning can enhance diagnostic accuracy for pediatric MS, potentially leading to earlier and more effective treatment. Clinicians should consider utilizing multimodal approaches in ambiguous cases to improve diagnostic confidence.
Conclusion
The study demonstrates the potential of multimodal deep learning models in enhancing the diagnostic process for pediatric MS using OCT. This approach may serve as a valuable tool in clinical practice for early identification and intervention.
