The introduction of population screening has resulted in largely increased detection of early stage breast cancers. A clinically challenging subcategory of such lesions are so-called ductal carcinoma in situ (DCIS). DCIS is histologically characterized by the proliferation of malignant epithelial cells that are bounded by the basement membrane of the ducts. DCIS lesions are noninvasive breast cancers, encompassing a wide spectrum ranging from low-grade harmless lesions to high-grade lesions with foci of invasive cancer. DCIS is considered a precursor for invasive ductal carcinoma (IDC). About 15 – 20% of patients with DCIS identified on core needle biopsy are under diagnosed, witnessed by detection of IDC upon breast conserving surgery (BCS) or mastectomy. At the same time, a large part (even up to 50%) of pure DCIS cases will never progress to invasive cancer if left untreated. Consequently, DCIS is associated with a large degree of over-diagnosis and overtreatment. Current diagnostic practice, which mainly consists of DCIS grading by a pathologist, is insufficiently capable of recognizing the malignant potential of these lesions. To reduce overtreatment without putting patients at increased risk, new biomarkers for improved risk stratification of DCIS are urgently needed.

It is a well-known phenomenon that malignant tumors are capable of transforming the connective tissue surrounding them (the so-called stroma). In this way they instigate a favorable environment, enhancing chances of metastatic spread while at the same time shielding the tumor from being attacked by the host immune system. Transformation of the stroma already starts in an early phase, and there is growing evidence that the events driving transformation of DCIS to invasion happen predominantly in the stroma rather than in tumor cells. We hypothesize that stromal alterations may serve as a proxy for the DCISs’ potential for invasive transformation, and that identification of such alterations will therefore yield strong prognostic information. We aim to use deep learning to create an algorithm that is capable of identifying the DCIS cases that have a potential for invasive transformation.




  • T. de Bel, M. Hermsen, J. Kers, J. van der Laak and G. Litjens, "Stain-Transforming Cycle-Consistent Generative Adversarial Networks for Improved Segmentation of Renal Histopathology", Medical Imaging with Deep Learning, 2019.