Development of a support system for breast diagnostic based on mammography data using deep learning

This work aims at developing an intelligent system to support breast diagnostics from mammographic image data. Some medical image databases are available to supply the data needed. The main objective is to help medical doctors' decisions for diagnostics. The literature shows that the tissues' densities can indicate possible malignancy, as its value varies for benign and malignant ones, due to the different Calcium molecules found in benign and in malignant tissues. The densities are related to the Hounsfield units, commonly used for mammogram image analyses by medical doctors. Thus, pattern recognition methods can be used to discriminate between benign and malignant tissues, supporting medical diagnostics based on mammograms. In particular, deep learning have showed to be very promising approaches for pattern recognition tasks in a variety of applications, especially for image processing. Thus, in this work, deep neural networks structures are investigated for the pattern recognition task, given mammogram image databases. Pre-processing methods are also investigated for image enhancement. Results are discussed. (author)