An approach to the prediction of breast cancer response to neoadjuvant chemotherapy based on tumor habitats in DCE-MRI images

For patients diagnosed with breast cancer who are undergoing neoadjuvant chemotherapy, predicting the response of the tumor to treatment is critical, as it enables the initiation of new therapies and improves patient care. In this context, dynamic contrast magnetic resonance imaging (DCE-MRI) is the most common imaging modality for assessing tumor response during breast cancer treatment. However, the visual inspection of a patient's tumor from DCE-MRI images by a specialist is a complex and repetitive task, due to the need to analyze high numbers of images, which makes the process exhausting and prone to error. Thus, this work presents a new methodology for predicting the response of a breast tumor to treatment, using four main steps: (i) segmentation of the tumor from DCE-MRI images of the breast; (ii) generation of tumor habitats from segmented DCE-MRI images; (iii) classification of the malignancy of tumor habitats; and (iv) prediction of the response of the breast tumor to treatment, based on the tumor habitats generated. The proposed methodology was tested on the public QIN Breast DCE-MRI dataset, and achieved a Dice score of 92.89% for tumor segmentation, an accuracy of 100% for the classification of malignancy, and an accuracy of 100% in terms of predicting the response of the tumor to the treatment applied. The results presented here demonstrate the efficiency of the proposed methodology, which can be integrated into a support system for treating patients with breast cancer.