To evaluate the effectiveness of multimodal deep learning models using Optical Coherence Tomography (OCT) for diagnosing pediatric onset multiple sclerosis (POMS) compared to non-inflammatory neurological conditions, emphasizing the critical need for early diagnosis.
Key Findings:
The early fusion model achieved the highest performance with an AUC of 0.90 and accuracy of 87%, indicating strong diagnostic capability.
Unimodal models performed well, with the best feature-based model yielding an AUC of 0.84, suggesting effective differentiation.
Late fusion models underperformed, achieving only 82% accuracy, highlighting the challenges in this approach.
Interpretation:
Multimodal learning with early fusion significantly enhances diagnostic performance by integrating spatial retinal information with clinically relevant structural features, capturing complementary patterns associated with MS pathology, which is crucial for accurate diagnosis.
Limitations:
The study's comparator group consisted of children with non-inflammatory neurological conditions, which may limit generalizability to broader pediatric populations.
The reliance on OCT data alone may overlook other relevant clinical information, potentially affecting diagnostic accuracy.
Conclusion:
The study demonstrates the potential of AI-driven tools in supporting pediatric neuroinflammatory diagnosis through enhanced classification accuracy using OCT data.
Patients with preoperative vitamin D deficiency had higher postoperative pain scores and opioid use after mastectomy, including more than triple the odds of moderate to severe pain within 24 hours of surgery.
