Interpretable Wavelet-CNN for Accurate Serum Raman Lung Cancer Diagnosis under Leakage-Safe, Patient-Level Splits.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Analytical Chemistry

Background:

• Clinical cancer diagnostics face challenges in distinguishing subtle molecular differences in serum.
• Machine learning (ML) models require interpretable feature attribution for clinical trust, especially with high sample similarity.
• Serum-based diagnostics are complex due to >95% shared chemical composition between healthy and diseased samples.

Purpose of the Study:

• To evaluate the accuracy and interpretability of a CWT-CNN deep learning model for clinical lung cancer diagnosis using serum Raman spectroscopy.
• To investigate the model's ability to handle biological variability in clinical samples.
• To identify specific spectral features driving diagnostic predictions.

Main Methods:

• Retrospective analysis of spontaneous Raman spectra from 213 patient serum samples (106 lung cancer, 107 controls).
• Application of Continuous Wavelet Transformation (CWT) combined with Convolutional Neural Networks (CNN) for spectral analysis and classification.
• Interpretability analysis using Gradient-weighted Class Activation Mapping (Grad-CAM) and inverse-CWT reconstruction.

Main Results:

• Achieved 90.5% accuracy in an independent validation cohort (19/21 correct diagnoses) with 91.7% sensitivity and 88.9% specificity.
• Demonstrated high performance using minimal sample volume (5 μL) and short acquisition time (10 min).
• Identified key spectral features at Raman shifts 1004 cm⁻¹ (phenylalanine), 1129 cm⁻¹ (lipid trans-conformation), 1458 cm⁻¹ (nucleotides), and 1560 cm⁻¹ (tryptophan) as crucial for classification.

Conclusions:

• The CWT-CNN approach maintains high diagnostic accuracy despite biological variability in clinical serum samples.
• The model provides biochemically meaningful feature attribution, linking spectral features to known cancer metabolism pathways.
• This data-first, interpretable ML approach shows promise for accurate and reliable clinical cancer diagnostics using Raman spectroscopy.