Label-free classification of breast cancer subtypes in ex vivo human tissues using Raman spectroscopy and machine learning

Breast conserving surgery (BCS) aims to excise breast tumors whilst preserving breast-related quality of life, but is complicated by the challenge of accurately identifying the margin between healthy and cancerous tissue. Raman spectroscopy (RS) has been shown to distinguish between normal breast tissue and breast cancer. Thus, this study aimed to further evaluate the diagnostic performance of RS in ex vivo breast tissue subtype classification via investigation of signals from healthy tissues and three breast cancer subtypes (invasive ductal carcinoma, IDC; invasive lobular carcinoma, ILC; and ductal carcinoma in situ, DCIS). A total of 80 tissue samples (46 normal and 34 cancerous) from 71 individuals were measured using a confocal Raman microscope. Spectral signatures were investigated, and supervised classification was performed for both two-class (healthy vs. cancer) and four-class (healthy vs. IDC vs. ILC vs. DCIS) classification tasks. RS successfully differentiated cancerous from normal breast tissue (97.84% sensitivity, 97.18% specificity). For four-class classification, RS achieved in-class sensitivity ranging from 83 to 96% and specificity from 93 to 99%. These findings demonstrate that RS can accurately distinguish normal from cancerous tissue and capture clinically relevant differences among histological subtypes, including invasive and pre-invasive disease, supporting its promise for intraoperative tissue characterization during BCS.