期刊名称：Indian Journal of Computer Science and Engineering
印刷版ISSN：2231-3850
电子版ISSN：0976-5166
出版年度：2021
卷号：12
期号：4
页码：1136-1148
DOI：10.21817/indjcse/2021/v12i4/211204222
语种：English
出版社：Engg Journals Publications
摘要：Breast Cancer is considered a significant health threat in women of all ages, and the incidence has been rapidly increasing, particularly in the last decade. Usually, for diagnosis, characteristics such as skin texture, color (redness), size variation, pain, and gene mutation are used. Binary classification is the most frequent classification type to classify benign and malignant tumors. A deep learning-based mechanism has been hugely popular and significant improvisation in binary classification, especially CNN-based mechanism. Moreover, to take CNN advantage, several breakthroughs have been observed for image classification, such as integrating layer-based features with the classifier. However, due to limited images available for medical research, several areas are exposed, such as overfitting issues, memory issues, and network architecture issues. Hence, in this research work, the design and development of an Improvised-DeepResidualNetwork, aka IDRN, for binary classification on mammogram. IDRN adopts the Deep Residual Network and enforces the improved transfer learning approach for training and testing the model. IDRN follows the novel architecture, forcing the transfer learning-based task-specific layer to be integrated to enhance the features and improvise the classification process. Further, IDRN is evaluated considering the INbreast dataset considering the important metrics like accuracy, precision, recall, F1-score, and AUC; a comparative analysis is also carried out with a state-of-art model of deep learning model.
