Multivariate analysis of breast tissue using optical parameters extracted from a combined time-resolved fluorescence and diffuse reflectance system for tumor margin detection

• Published in: Journal of biomedical optics Vol. 28; no. 8; p. 85001
• Main Authors: Dao, Erica, Gohla, Gabriella, Williams, Phillip, Lovrics, Peter, Badr, Fares, Fang, Qiyin, Farrell, Thomas J., Farquharson, Michael J.
• Format: Journal Article
• Language: English
• Published: Bellingham S P I E - International Society for 01-08-2023; Society of Photo-Optical Instrumentation Engineers
• Subjects: Breast cancer; Cancer; Cancer therapies; Classification; Datasets; Decision analysis; Decision making; Decision trees; FDA approval; Fluorescence; Investigations; Lasers; Mathematical models; Modelling; Multivariate analysis; Optical properties; Parameters; Photonics; Prediction models; Principal components analysis; Reflectance; Sensitivity; Surgeons; Tissue analysis; Tissues; Tumors; United States > US
• ISSN: 1083-3668; 1560-2281
• DOI: 10.1117/1.JBO.28.8.085001

SignificanceBreast conservation therapy is the preferred technique for treating primary breast cancers. However, breast tumor margins are hard to determine as tumor borders are often ill-defined. As such, there exists a need for a clinically compatible tumor margin detection system.AimA combined time-resolved fluorescence and diffuse reflectance (TRF-DR) system has been developed to determine the optical properties of breast tissue. This study aims to improve tissue classification to aid in surgical decision making.ApproachNormal and tumor breast tissue were collected from 80 patients with invasive ductal carcinoma and measured in the optical system. Optical parameters were extracted, and the tissue underwent histopathological examination. In total, 761 adipose, 77 fibroglandular, and 347 tumor spectra were analyzed. Principal component analysis and decision tree modeling were performed using only TRF optical parameters, only DR optical parameters, and using the combined datasets.ResultsThe classification modeling using TRF data alone resulted in a tumor margin detection sensitivity of 72.3% and specificity of 88.3%. Prediction modeling using DR data alone resulted in greater sensitivity and specificity of 80.4% and 94.0%, respectively. Combining both datasets resulted in the improved sensitivity and specificity of 85.6% and 95.3%, respectively. While both sensitivity and specificity improved with the combined modeling, further study of fibroglandular tissue could result in improved classification.ConclusionThe combined TRF-DR system showed greater tissue classification capability than either technique alone. Further work studying more fibroglandular tissue and tissue of mixed composition would develop this system for intraoperative use for tumor margin detection.