Purpose: Breast cancer (BC) is the most common cancer in women worldwide. There exist various advanced chemotherapy drugs for BC; however, chemotherapy drugs may result in brain damage during treatment. When a patient's brain is changed in response to chemo drugs, it is termed chemo-brain. In this study, we aimed to construct machine-learning models to detect the subtle alternations of the brain in postchemotherapy BC patients.
Methods: Nineteen BC patients undergoing chemotherapy and 20 healthy controls (HCs) were recruited for this study. Both groups underwent resting-state functional MRI and generalized q-sampling imaging (GQI).
Results: Logistic regression (LR) with GQI indices in standardized voxel-wise analysis, LR with mean regional homogeneity in regional summation analysis, decision tree classifier (CART) with generalized fractional anisotropy in voxel-wise analysis, and XGBoost (XGB) with normalized quantitative anisotropy had formidable performances in classifying subjects into a chemo-brain group or an HC group. Classifying the brain MRIs of HC and postchemotherapy patients by conducting leave-one-out cross-validation resulted in the highest accuracy of 84%, which was attained by LR, CART, and XGB with multiple feature sets.
Conclusions: In our study, we constructed the machine-learning models that were able to identify chemo-brains from normal brains. We are hopeful that these results will be helpful in clinically tracking chemo-brains in the future.
Keywords: breast cancer (BC); chemo-brain; generalized q-sampling imaging (GQI); machine learning; resting-state functional magnetic resonance imaging (rs-fMRI).
© 2018 International Society for Magnetic Resonance in Medicine.