Hyperspectral Imaging for Benign and Malignant Diagnosis of Breast Tumors.

Objective: To assess the feasibility of combining microscopic hyperspectral imaging (370-1100 nm) with a lightweight 1D-CNN for rapid, label-free discrimination of benign and malignant breast tumors.

Methods: Breast specimens (43 malignant, 39 benign) were imaged; 2 050 000 pixel spectra were preprocessed (dark-current subtraction, white-reference calibration, Savitzky-Golay smoothing, z-score normalization) and input to a custom 1D-CNN. Performance was benchmarked against SVM, AlexNet, and LSTM using accuracy, sensitivity, specificity.

Results: The 1D-CNN achieved 90.43% accuracy, 89.10% sensitivity, 91.34% specificity, exceeding baseline models.

Conclusions: Combining HSI with 1D CNN enables rapid and highly accurate classification of breast tumors, providing a new approach to rapid pathological diagnosis.