Abstract
Breast cancer is the most prevalent cancer among women worldwide, emphasizing the need for rapid and accurate diagnostic tools to improve patient outcomes and survival rates. In this study, we developed a diagnostic tool-a multispectral pen based on diffuse reflectance spectroscopy (DRS)-to enable real-time ex vivo differentiation between malignant and adjacent normal human breast tissues, primarily based on lipid and collagen absorbers. Diffuse reflectance was observed to be higher, while reduced absorbance was lower for malignant tissue compared to adjacent normal tissue across 62 samples (50 formalin-fixed and 12 fresh tissues). Clinical data from 31 patients, with paired adjacent normal and malignant samples per patient, revealed significantly lower mean reduced absorbance values for malignant tissue at three wavelengths: 850 nm (0.13 ± 0.02 vs. 0.28 ± 0.02), 940 nm (0.19 ± 0.01 vs. 0.37 ± 0.02), and 1050 nm (0.25 ± 0.03 vs. 0.43 ± 0.03) with p < 0.0001 across formalin-fixed tissue samples. Using a CatBoost machine learning model, the tool achieved an accuracy of 90%, a sensitivity of 80%, and a specificity of 100% in distinguishing malignant from normal tissues. Limitations include a small sample size and limited patient diversity. Future work aims to integrate the DRS technology with machine learning algorithms to provide real-time intraoperative margin assessment, and testing in a larger study.