Tumor Segmentation in Breast MRI Using Deep Learning

An uncontrolled growth of tissues in a body part and forming a mass that can either be benign or malignant is referred to as a tumor. Early screening and detection of such a mass are crucial for the best prognosis. In the past five years, 7.8 million women have survived breast cancer, making it the most prevalent cancer in females. A medical imaging method such as MRI or mammography is the first step in a clinical diagnosis. Afterward, a biopsy is the next step if the imaging test shows tissue changes. Consequently, a patient's prognosis is determined by the analysis of the MRI. RIDER Breast MRI data collection is collected hosted on The Cancer Imaging Archive. Our aim is to develop a deep learning method to segment the masses in 2D breast MR images, axial view. Furthermore, an in-depth investigation into the importance of TL (Transfer Learning) will be conducted. A goal of the project is to build and train the U-net model from scratch on the dataset, then test the dataset on the U-net model using TL. When TL is used, the models like Inception and VGG without the last dense layers are used as the encoder part of the network for feature extraction.