Based on Raman spectroscopy and clinical laboratory tests: Developing a fusion model to predict neoadjuvant therapy response in breast cancer.

BACKGROUND: Neoadjuvant therapy (NAT) constitutes a pivotal component of comprehensive breast cancer treatment. Early prediction of NAT efficacy is crucial for personalised treatment and improving patient prognosis. However, effective early prediction methods remain lacking at present. METHODS: A retrospective cohort study was conducted involving 406 patients with invasive breast cancer. Baseline clinical laboratory parameters and serum Raman spectra were collected. Key features were selected using an attention mechanism to construct a multimodal fusion prediction model based on Transformers. The model was trained and its predictive efficacy evaluated. RESULTS: The mean age was 50.88 years. Overall, 134 women achieved pCR following NAT. The fusion model outperformed single-modality approaches in predicting pCR, achieving the highest AUC (0.790) and accuracy (0.752). This performance significantly surpassed single-modality models based solely on clinical laboratory tests (AUC = 0.759) or Raman spectroscopy (AUC = 0.669). Key predictive features identified include lipase, Ki-67 index, HER2 status, HFR%, and albumin. CONCLUSION: Multimodal fusion of Raman spectroscopy with clinical laboratory indicators effectively enhances the predictive performance of NAT efficacy, providing a non-invasive, dynamic adjunctive decision-making tool for personalised breast cancer treatment.