Deep Learning-Assisted Differentiation of Four Peripheral Neuropathies Using Corneal Confocal Microscopy

Area of Science:

• Ophthalmology
• Neurology
• Artificial Intelligence

Background:

• Peripheral neuropathies cause significant patient disability.
• Current diagnostic methods are limited, leading to late or missed diagnoses.
• Specialized testing facilities are often inaccessible.

Purpose of the Study:

• To develop an interpretable deep learning system, NeuropathAI, for rapid, automated diagnosis of peripheral neuropathies.
• To differentiate between diabetic peripheral neuropathy (DPN), chemotherapy-induced peripheral neuropathy (CIPN), chronic inflammatory demyelinating polyneuropathy (CIDP), and HIV-associated sensory neuropathy (HIV-SN).

Main Methods:

• A deep learning-based multiclass classification system was developed.
• The system analyzes corneal nerve images for diagnosis.
• Model performance was evaluated using AUC, F1-score, sensitivity, specificity, and precision.

Main Results:

• NeuropathAI achieved high accuracy with an overall AUC of 96.75%.
• The system demonstrated excellent discrimination for CIDP (97% AUC), CIPN (93.1% AUC), HIV-SN (99.7% AUC), and DPN (96.9% AUC).
• Heatmap visualizations confirmed model interpretability by localizing to areas of nerve fiber loss.

Conclusions:

• NeuropathAI enables rapid and accurate diagnosis of four prevalent peripheral neuropathies.
• AI-driven corneal image analysis shows significant potential for diagnosing and differentiating peripheral neuropathies.
• This technology addresses a critical clinical gap in diagnosing peripheral neuropathies.