Artificial Intelligent Label-Free SERS Profiling of Serum Exosomes for Breast Cancer Diagnosis and Postoperative Assessment.

Abstract

Breast cancer subtypes have important implications of treatment responses and clinical outcomes. Exosomes have been considered as promising biomarkers for liquid biopsies, but the utility of exosomes for accurate diagnosis of distinct breast cancer subtypes is a grand challenge due to the difficulty in uncovering the subtle compositional difference in complex clinical settings. Herein, we report an artificial intelligent surface-enhanced Raman spectroscopy (SERS) strategy for label-free spectroscopic analysis of serum exosomes, allowing for accurate diagnosis of breast cancer and assessment of surgical outcomes. Our deep learning algorithm trained with SERS spectra of cancer cell-derived exosomes is demonstrated with a 100% prediction accuracy for human patients with different breast cancer subtypes who do not undergo surgery using SERS spectra of serum exosomes. Furthermore, when combined with similarity analysis by principal component analysis, our approach is able to evaluate the surgical outcomes of breast cancer of distinct molecular subtypes.