Early Detection of Breast Cancer by Deep Learning Algorithms through Histopathological Images
Raghvendra Kumar, Kanchi Mishra, Suhani Dhiraj Sunehra · 2024
Breast cancer is a significant global health concern, affecting both women and men, with early detection playing a critical role in improving prognosis. Invasive ductal carcinoma, a typical type of breast cancer, accounts for approximately 80% of cases. Histopathological images provide microscopic views of stained tissue samples, offering detailed insights into cellular structures and aiding in the identification of cancerous regions. This research paper aims to revolutionize breast cancer interpretation with early recognition of breast cancer using Deep Learning (DL) algorithms with datasets collected from Kaggle. An image generator is utilized to enhance the quality and diversity of the data, including noise reduction, contrast enhancement, and image normalization. These techniques improve feature visibility and reduce artifacts, ensuring the accuracy of subsequent analyses. Machine learning (ML) techniques, particularly Convolutional Neural Networks (CNNs), are applied to the preprocessed histopathological images for breast cancer detection. Transfer learning is incorporated using well-established CNN architectures, such as ResNet-50 and InceptionV3, to leverage their learned features and optimize performance. In this research paper, a comparative study is conducted to assess the performance of CNNs using ResNet-50 and InceptionV3 in detecting Invasive Ductal Carcinoma (IDC). The baseline CNN produced an accuracy of 93.4%, the ResNet-50 and InceptionV3 models produced an accuracy of 84% and 91.2% respectively. This research aims to reduce the reliance of healthcare professionals on subjective assessments, improving diagnostic workflow, and ultimately assisting healthcare professionals in making more precise and timely decisions.