Breast Cancer Determination Using B-Mode Ultrasound Imaging Using Deep Learning Technique

This research evaluates multiple deep-learning architectures for predicting breast cancer. The study meticulously examines the performance of prominent convolutional neural network (CNN) models, including VGG16, ResNet50, InceptionV3, SqueezeNet, and Xception, in accurately identifying breast cancer. Preprocessing techniques are integrated to standardize the dataset's image dimensions and pixel values, which are comprised of three classes: normal, benign, and malignant. The training and validation procedures are meticulously executed, considering critical performance metrics such as accuracy and loss. This paper offers comprehensive insights into each model's suitability for real-world breast cancer prediction. Notably, the analysis underscores the Xception architecture's promising performance and innovative design, advocating its adoption in predicting breast cancer. By shedding light on the capabilities of these deep learning architectures, this study significantly contributes to advancing our understanding of their real-world utility in image classification tasks. Furthermore, it serves as a valuable resource for guiding model selection across various domains, including critical areas like healthcare, where accurate image classification is paramount. Drawing upon a diverse dataset curated from Kaggle, this research transcends mere accuracy enhancement, emphasizing the necessity of considering multifaceted factors when choosing deep learning models for real-world deployment. Xception leading with a notable training accuracy of 98.56% and a validation accuracy of 96.48%. Through this exhaustive evaluation, the paper underscores the importance of informed decision-making. It highlights the transformative potential of deep learning architectures in addressing complex image classification challenges across diverse domains.