Multimodal machine learning for distinguishing pediatric multiple sclerosis from non-inflammatory conditions using...

Area of Science:

• Ophthalmology
• Neurology
• Artificial Intelligence

Background:

• Early diagnosis of pediatric multiple sclerosis (MS) is critical for effective therapeutic intervention.
• The anterior visual pathway is a key diagnostic consideration for MS.
• Optical coherence tomography (OCT) provides high-resolution retinal imaging, reflecting structural integrity.

Purpose of the Study:

• To determine if multimodal deep learning models can diagnose pediatric onset MS (POMS) using only OCT data.
• To evaluate the diagnostic performance of different deep learning approaches for POMS detection.

Main Methods:

• Analysis of 3D OCT scans from children with POMS and controls.
• Evaluation of three classification approaches: deep learning with classical ML, ML on OCT features, and multimodal fusion (early and late).
• Inclusion of raw macular and optic nerve head images, and 52 automatically segmented features.

Main Results:

• The early fusion multimodal model achieved the highest performance (AUC: 0.90, weighted F1: 0.87, accuracy: 87%).
• This multimodal approach outperformed unimodal feature-based and image-based models.
• Late fusion models showed lower diagnostic accuracy, particularly for the minority class.

Conclusions:

• Multimodal learning with early fusion significantly enhances diagnostic performance for POMS.
• This AI-driven approach effectively combines spatial retinal information and structural features.
• The findings suggest a promising AI tool to aid in pediatric neuroinflammatory diagnosis.