Predicting the efficacy of neoadjuvant chemotherapy for breast cancer based on dynamic diffuse spectral imaging

Objective: Neoadjuvant chemotherapy (NAC) is a common treatment for breast cancer patients while its efficacy shows high heterogeneity. Currently, there is a lack of non-invasive method to accurately predict the treatment outcome early. This study is based on dynamic diffuse optical spectroscopy (DOS) imaging combined with a dynamic pressure modulation mechanism to quantify tissue blood oxygen level, and aims to achieve early and accurate prediction of NAC efficacy. Methods: A total of 268 NAC patients were enrolled in this study. Dynamic spectral images at dual wavelengths (640/808 nm) were collected using the DOBI ComfortScan® system under stepwise pressure loading (5→10→5 mmHg). Four pressure-time points (first low-pressure endpoint, high-pressure start point, high-pressure endpoint, and second low-pressure start point) were defined to construct standardized ratio maps Ri. Ratio-of-ratio images (RRi) were generated through dual-wavelength results and were subsequently converted into oxygen saturation maps (SO₂i). Average SO₂ features in tumor (T) and non-tumor (N) regions were extracted. Using pathological Miller-Payne grading (G1-G2 for insensitive efficacy, G3-G5 for sensitive efficacy) as the gold standard, machine learning algorithms including Random Forest (RF) and Support Vector Machine (SVM) were used to construct prediction models. SMOTE oversampling and five-fold cross-validation were adopted to optimize the performance. Results: Early predictability: From before the first chemotherapy to after the first chemotherapy, the RF model achieved a sensitivity of 85.71% (accuracy 77.27%, specificity 75.68%, AUC 0.82) using SO₂1 in the tumor region for predicting MP grades G1-G2. Optimal performance: From before the first chemotherapy to before the third chemotherapy, the RF model achieved an accuracy of 90.32% (sensitivity 80.00%, specificity 92.31%, AUC 0.85) using SO₂4 in the non-tumor region (N). Conclusion: This study demonstrates the potential of dynamic SO₂ features as non-invasive biomarkers for early prediction of NAC efficacy, which is expected to provide evidence-based basis for early clinical adjustment of treatment plans. In the future, we will expand the sample size and integrate molecular phenotype information to verify the method, and optimize the multivariate prediction models by integrating clinical factors, with the ultimate goal of promoting the translational application of diffuse optical technology in individualized tumor treatment.