Résumé
Background and purpose: Side-effects after breast cancer treatment, including radiotherapy (RT), can drastically affect a patient's quality of life. This study aims to identify anatomical regions where RT dose is related to breast side-effects using the voxel-wise image-based data mining (IBDM) technique. Materials and Methods: Data of 922 patients included in the REQUITE study and who underwent RT after breast-conserving surgery were analysed. All patients were treated supine. We analysed breast pain, oedema, atrophy, induration, and nipple retraction, scored before and after RT by clinicians and patients. We investigated the toxicity burden using the area under the curve for toxicity over time (ToxT-AUC) and applied IBDM by normalising all dose distributions to three reference anatomies. To do so, we established a spatial normalisation pipeline, ensuring that the ipsilateral breast was aligned (flipping all left-side treatments laterally). Dose distributions were converted to equivalent dose in 2 Gy-fraction (EQD(2), alpha/beta = 1.7 Gy and 3 Gy) to account for different RT fractionations. To determine the relevance of each identified region, we used multivariable ordinal regression, including patient-specific factors and clinical variables. Results: Significant regions associated with each breast side-effect were found in different breast quadrants. The regions showed higher mean EQD(2) than the complete breast contour, particularly for grade 2+ (p < 0.01). Dosimetric parameters extracted from the region were significant in multivariable analysis. Conclusion: We found regions within the breast where the local dose associates with breast side-effects after breast RT. This result opens a new hypothesis which could help reduce side-effects after breast radiotherapy.