Optical Breast Spectroscopy for Personalized Early Cancer Risk
Abstract
Purpose
This study aimed to develop, validate, and corroborate an innovative Breast Asymmetry Score (BAS) derived from optical breast spectroscopy (OBS) as a non-invasive indicator of breast cancer risk, and to integrate BAS with established epidemiological risk factors using artificial neural networks. This will improve risk stratification in women younger than 40 years.
Methods
A total of 23,479 individual OBS scans were obtained, from which 1,198 such patients were selected who met the desired selection criteria. Using a 13-wavelength near-infrared spectroscopy system operating between 660 and 1050 nm, bilateral breast measurements were performed. Using spectral decomposition, concentrations of key chromophores, including water, collagen, hemoglobin, and lipid, were extracted. BAS was computed as the Euclidean distance between bilateral differences in tissue composition. Optical Breast Density (OBD) was quantified on a standardized 0–100 scale reflecting fibro glandular content. Missing spectral data at 915 nm were interpolated from adjacent wavelengths, with validation thus, confirming high spectral continuity. NHIS and PLCO data sets, incorporating demographic, familial, hormonal, and reproductive variables were used to train deep neural network models. Model performance was assessed using five-fold cross-validation and area under the ROC curve.
Results
BAS exhibited a right-skewed distribution, identifying a high-asymmetry subgroup, while OBD demonstrated a bimodal distribution, with 42% classified as dense breast tissue. Lipid and hemoglobin content showed strong bilateral correlations. Neural network models attained training AUC values of approximately 0.75 and validation AUC values around 0.73. Sensitivity of 79.3%, specificity of 55.3%, and negative predictive value of 97.8%, were obtained at optimal thresholds.
Conclusion
OBS-derived BAS offers a quantitative measure of bilateral breast tissue asymmetry. When combined with conventional risk factors, BAS determines clinically acceptable predictive performance, supporting its potential role as a non-invasive breast cancer risk assessment tool for women aged 18–40 years not eligible for routine mammographic screening.