3D Segmentation and Subtypes Classification of Breast Cancer Using Ultrasound Images Through Hybrid Approach of CNN and LSTM

Since cancer in the breast is a common and potentially fatal illness, early and precise identification is essential to enhancing patient outcomes. The analysis of medical images, especially ultrasound imaging, has shown great potential for deep learning approaches in recent years. In order to segment and identify malignancies in all three dimensions from ultrasound pictures, this novel hybrid approach combines long short-term memory (LSTM) networks with convolutional neural networks (CNN). Our novel approach enhances the efficacy and precision of breast tumor diagnosis by fusing the strength of Long Short-Term Memory with Convolutional Neural Networks”. The 3D segmentation component’s multi-atlas registration methodology offers a thorough way to accurately outline the tissues of the breast in space. Ultrasound images are utilized to train the model. These images contain subtle patterns that may indicate clinical problems. Additionally, our system employs the Super Pixel segmentation approach to help identify regions of interest to more precise tissue abnormality detection. Our proposed approach aims to offer a robust basis for 3D segmentation and breast cancer diagnosis by symbiotically combining the features of CNNs & LSTMs in this hybridization structure.