<p>Achieving complete resection of breast cancer with clear margins remains a significant surgical challenge, particularly for infiltrating subtypes, where re-resection rates of up to 45% have been reported. Consequently, a device capable of providing real-time feedback to surgeons regarding the resected breast tissue holds the potential to significantly improve R0 resection rates. This study represents a further advancement toward intraoperative, real-time classification of breast tissue using optical emission spectroscopy (OES). The objective was to establish the feasibility of OES for distinguishing between normal and pathological breast tissue during electrosurgical incision. Spectra obtained from specimens of 80 patients were analyzed, including 68 patients who underwent breast cancer surgery and 12 patients who underwent risk-reducing or breast reduction surgery. Spectroscopic classification was performed of spectra from tumors with no special type (NST) invasive-lobular carcinoma (ILC) using a machine learning approach based on selected spectral features. The true positive rate reached 91,0% for NST spectra and 78.7% for ILC spectra with true negative rates of 95,2% for NST and 85,4% for ILC. In summary, the current support vector machine (SVM) algorithm demonstrates reliable classification performance for the predominant NST subtype. However, the accuracy of the ILC classification still needs to be improved. Further refinement of the OES-based classification approach is necessary to enhance its reliability across all breast cancer subtypes, particularly the rarer forms, thereby facilitating robust real-time detection during surgery.</p>