43007 Tarragona, Spain
This project will be directed to the detection of local recurrence (probability of relapse) through the quantitative changes perceived in breast density (fibrosis) after applying different radiotherapy techniques (hypofractionation or control mammography). Besides, we will investigate the correlation of such quantification as a potential measure to predict genetic susceptibility to adverse effects of radiotherapy using DNA samples from selected cases. This will be performed using the following steps: (1) Developing a DL model based on a conditional GAN to implement dense tissues segmentation in mammographic images (i.e., 2D mammography or pseudo-3D tomosynthesis). (2) Developing a multi-class CNN architecture for breast density classification using the resulted segmented masks based on BI-RADS standard (i.e., fatty, scattered fibroglandular, heterogeneously, and extremely). (3) Constructing a fully automatic method for breast density estimation. (4) Improving breast density estimation that might be achieved by using alternative breast imaging, such as MRI. (5) Studying the relationship between breast density, the genetic susceptibility to BC relapse and the effects of radiotherapy treatment. (6) a fully automatic follow-up of the quantification of breast density of multiple temporal mammograms for estimating the risk of the relapse. We will use the preliminary work of our partner URV fully automated breast density segmentation and classification using DL as preliminary work for breast density estimation.
As a result, 1) a biomarker will be obtained from individual 2D mammography or pseudo-3D tomosynthesis as well as MRI, to provide a volumetric breast density estimator for a set of images. 2) Evolutionary models will also be obtained by means of data mining techniques and automatic learning of health episodes, starting from the changes in breast density over time (physiology) and the response to treatment by the patients (quality). 3) These two models will be used to perform toxicity studies and make medium- and longterm breast tissue degradation forecasts customized for each patient, thus anticipating the relapse and local recurrence. As a result, corrective actions (radiotherapy dose adjustments) can be undertaken to minimize secondary effects.