In Vitro Detection of Breast Cancer Cell Types Using Machine Learning-Assisted Spectral Fingerprinting of SWCNTs

Abstract

The early detection of breast cancer currently relies on expensive mammography, followed by pathology that uses biopsied, fixed, and immunohistochemically stained tissues. A live-cell detection approach could be highly beneficial as a supportive diagnostic and research tool to better understand and resolve the dynamic nature of breast cancer cells and their response to treatment in real time. Here, we present a single-walled carbon nanotube (SWCNT) near-infrared fluorescence spectral fingerprinting approach combined with machine learning to precisely detect the heterogeneity of breast cancer cells in live culture. We introduced DNA-functionalized SWCNTs to MCF-10A (a non-tumorigenic healthy control) and cancer cell lines spanning known extrinsic disease subtypes: MCF-7 (luminal A), HCC1954 (HER2+), MDA-MB-231, and MDA-MB-468 (both triple-negative). The NIR fluorescence spectra of DNA-SWCNTs across 600 individual cells within each type showed significant differences in emission peak intensities, center wavelengths, and peak intensity ratios, attributable to variations in cellular uptake and biomolecular interactions. These spectral changes likely arise from complex SWCNT-cellular interaction "fingerprint" that includes redox-mediated modulation of the local nanotube environment, rather than from a single biomarker response. The extracted spectral features were used to train an ensemble machine learning model. The model achieved 98% classification accuracy for breast cancer detection and 95% classification accuracy for breast cancer cell subtyping. Moreover, Raman microscopy further showed that MDA-MB-468 cells exhibited the highest SWCNT uptake, whereas MCF-10A cells showed greater SWCNT aggregation, consistent with their lower broadband NIR fluorescence intensity. These results demonstrate that SWCNT NIR fluorescence fingerprints can capture cell line-specific optical signatures. This platform provides a foundation for nanomaterial-enabled biosensing strategies aimed at real-time monitoring of cancer-associated cellular states.