Batch effect correction for LIBS-FTIR spectral fusion in breast cancer serum detection based on gradient reversal...

Breast cancer is the most prevalent malignancy among women, necessitating rapid, non-invasive, and cost-effective diagnostic technologies for early detection. Spectroscopic techniques like laser-induced breakdown spectroscopy (LIBS) and Fourier transform infrared spectroscopy (FTIR) have shown promise in the area. However, batch effects caused by factors like operator variability and instrument fluctuations can compromise the generalization ability of trained models across different sample batches. In this study, we combined LIBS and FTIR to analyze breast cancer serum samples from different batches and developed a Gradient Reversal Adversarial Network (GRAN) to correct batch effects. The GRAN model incorporated an adversarial mechanism between a batch classifier and a label classifier, enabling it to learn batch-invariant features. The results demonstrated that GRAN substantially enhanced classification accuracy across different batches, achieving a test accuracy of 89.7 % when trained on one batch and tested on another. This represents a significant improvement over traditional models, which achieved test accuracies of 63.7 % under comparable conditions. The findings of this research lay the foundation for a robust and non-invasive breast cancer detection method by leveraging the GRAN model for spectral fusion.