Mitigating Diagnostic Disparities in Melanoma Screening: A Robust 3D Spatial-Spectral Deep Learning Framework for Resource-Limited Settings

Melanoma is the most fatal skin cancer, yet early diagnosis is severely hampered by diagnostic health disparities. In resource-constrained settings, reliance on subjective visual inspection and limited access to specialized pathology often leads to high rates of missed malignancies (False Negatives), contributing significantly to health inequities in cancer care. Hyperspectral Imaging (HSI) offers detailed biochemical data that can mitigate this subjectivity.

We developed an integrated deep learning framework for Melanoma (MEL) versus Nevus (NEV) differentiation using HSI data. The methodology involves a robust preprocessing pipeline, including data augmentation for sample size mitigation. The key innovation is the use of Randomized Singular Value Decomposition (RSVD) for efficient spectral feature extraction, feeding a customized 3D Convolutional Neural Network (3D CNN) architecture to synchronously analyze the spatial-spectral data. The model was evaluated on a balanced test set (n=100: 50 MEL, 50 NEV) from the histology-verified Hyperspectral Dermoscopy dataset.

The RSVD-3D CNN model demonstrated exceptional diagnostic performance on the test set, achieving an overall Test Accuracy of 99.00%. Critically, the model achieved a perfect 1.00 Sensitivity (Recall) for the Melanoma class, resulting in zero False Negative predictions. The training and validation loss curves converged stably at a low value (Val Loss $\approx 0.03$), confirming model robustness and generalizability.

This deep learning framework sets a new benchmark for HSI-based melanoma diagnosis by achieving clinical-grade reliability. The ability to eliminate False Negatives directly addresses a critical diagnostic disparity in resource-limited settings. The model serves as an essential, non-invasive Implementation Science tool that can be deployed to provide objective triage, significantly improving health equity and ensuring timely access to life-saving care for underserved populations.