MammoLearning Project

Breast cancer is one of the most common forms of cancer worldwide, and the most common type of cancer for women. It is estimated that 1 in 8 women will experience some form of breast cancer worldwide. However, in recent decades there has been a gradual reduction in deaths from this type of cancer. This trend can be attributed to the extensive and regular use and technological improvement of the mammography diagnostic technique and, in particular, to the early identification of cancerous tumors. The Medical Center of the Ormylia Foundation, in the context of its social and charitable work, has been offering free preventive examinations for the early detection of two cancer types, cervical cancer and breast cancer for two decades. The Center provides the possibility of chronic monitoring of patients with simultaneous high-level medical supervision, using the most modern laboratory and software methods. It is estimated that 23,680+ women have been examined so far, with a review rate of 93.2%. Every year 20-25 breast cancers are diagnosed with 87% of the cancers being in an early, curable stage.

The purpose of this project will be the implementation of a deep learning algorithm to identify breast tumours and the possible prediction of primary carcinogenesis to optimize the therapeutic process.

Initially, the deep learning algorithm will be implemented and will be trained with the help of the rich database of the Medical Diagnostic Center of the Ormylia Foundation. In particular, the effort will focus on the images of mammograms from the tens of thousands of women tested, which have been examined in the Diagnostic Center in recent decades*. The existence of such a volume of data is capable of initializing the training of the algorithm.

In the second phase, simulation data will be used for both education and control of the deep learning algorithm, using the Radiographic Imaging Simulation System. This system has been developed in the Biomedical Engineering Unit of the University of Patras and is already used by many research teams worldwide. It consists of modules that allow the design of breast models and the formation of mammogram imaging images. The ability to produce large data (Big Data) by simulation, coupled with the fact that there is an a priori knowledge of existence or not, exact location and structure of cancerous lesions, offers the opportunity to optimize the learning procedure and control process of the deep learning algorithm.

*It is obvious and necessary to maintain full anonymity as to the identities of patients assigned to mammography..