Pre-operative classification of Ductal Carcinoma in Situ : a deep learning method.

This thesis presents a new and innovative framework for the detection and classification of clusters of calcifications in mammographic images. The study shows that characteristic associations can be found between the appearance of DCIS lesions on mammography and the underlying behaviour and/or risk of progression towards IBC. The proposed deep learning-based system is capable of addressing the detection and classification of calcifications in mammographic images simultaneously. If put into context with additional clinical information, the findings presented in this study could potentially help navigate pre-biopsy discussions with patients. Better knowledge about the differences in DCIS lesions and behaviour can aid both patients and clinicians to make better-informed evidence-based decisions regarding possible treatments options, including the option to refrain from treatment.