Abstract

Breast Cancer (BC) is the most diagnosed and the second most frequent cause of death among women. Despite the wide application of mammography as a screening technique, the low sensitivity of this technique in dense breasts, and the risks of radiation are significant disadvantages. In diagnosing and early treatment planning of breast cancer, medical imaging is very important. This paper provides a framework for the segmentation and 3D reconstruction of breast tumors based on a 3D NIFTI image dataset. The preprocessing step entails using an integration of Non-Local Means (NLM) filter and Sobel edge detector to effectively minimize noise and maintain significant structure of the breast tissue. Among the three deep learning models, including UNet, ResNet, and a hybrid architecture of UNet-ResNet, the segmentation of tumor regions is carried out. These models are tested on traditional measures such as Dice Coefficient and Jaccard Index, with the hybrid UNet-ResNet model performing better than others, with a Dice coefficient of 1.38 and a Jaccard index of 0.08. After the segmentation, the volume of the tumor is estimated through the spatial dimensions of the tumor, and the Marching Cubes algorithm is used to create a 3D image of the cancerous tissue to be easily visualized. To prevent the invasion of patient privacy and data security, the system will involve encryption-based storage, anonymization, and secure access control, which are in line with healthcare data protection. This framework not only enhances the accuracy of tumor segmentation but also ensures safe handling of sensitive MRI data, which eventually helps in making better clinical decisions.