Date of Publication :28th June 2024

Abstract:Breast cancer is a significant health concern among women across the globe. Detecting cancerous cells in their early stages is challenging for oncologists and radiologists due to the asymptomatic nature of the disease, making timely treatment difficult. Common traditional approaches like mammography, ultrasound, MRI, and CT scans often give inaccurate results with false positives and negatives due to errors, affecting the accuracy of the diagnosis. This research proposes an innovative approach focusing on the automated ultrasound image segmentation and classification of breast cancer with Attention U-Net and Convolutional Neural Networks framework. Iterative increments of precision and accuracy enabled the data to dynamically adjust and acquire knowledge from developing masked image patterns. We used the "Breast Ultrasound Images Dataset" to improve our model's ability to analyse and understand breast ultrasound images, helping us make advancements in medical imaging research and applications. The dataset categorizes images into normal, benign, and malignant classes, with corresponding ground truth images. The developed model exhibits a remarkable accuracy of 93.3% in efficiently and accurately identifying and categorizing various types of noise and systematic errors. Its exceptional capability contributes to the facilitation of timely and effective treatment.

Reference :