Considering the growing number of breast cancer cases and people being screened for breast cancer yearly, we proposed an integrated end-to-end deep learning pipeline for Breast Ultrasound Image Segmentation leading to Classification that would support medical practitioners with an easily adoptable computer-aided automated framework for an early diagnosis of breast cancer. In this study, we successfully adopted the biomedical and cancer research domain at every stage of our framework in terms of analyzing breast ultrasound images in the most effective way. However, to tackle the highly noised Breast Ultrasonography Images with complex artifacts that possess inter-class similarity and multi-co-linearity issues, we have preprocessed the Breast Ultrasonography Images with Simple Linear Iterative Clustering (SLIC) followed by supervised segmentation with modified U-Net architecture. For the classification module, we have utilized a lightweight neural network, integrated with pretrained transfer learning model for feature extraction followed by the well designed fully connected network. This study also explores the challenges of preprocessing the Breast Ultrasound Images with unsupervised image segmentation methods including K Means ++ and SLIC. To validate the efficiency of our proposed automated pipeline, we have experimented with our models on a very challenging Breast Ultrasonography Image Dataset and obtained outperforming results.
