Enhanced machine learning models for accurate breast cancer mammogram classification

Abstract

Breast cancer has now become the leading cancer type in Indian urban regions surpassing cervical cancer. The imperfect nature of current diagnostic techniques calls for more dependable assessment methods because new automated diagnostic systems do not totally eliminate imperfections. The research of Indian breast cancer datasets remains lower than international platforms owing to distinctive features of breast density, texture, lesion size and composition between these populations. The research develops automated breast cancer detection algorithms for Indian breast types that incorporate metadata to enhance system accuracy in medical diagnosis. We created machine learning algorithms for mammography imaging as part of a development process optimized for Indian breast cancer features. Our top-performing individual model achieved an AUC of 0.95 per image. When integrating four models, the AUC increased to 0.98, with independent test set results from the INbreast database showing 86.7 % sensitivity and 96.1 % specificity. These findings highlight deep learning's potential to enhance mammographic assessment by improving diagnostic accuracy, reducing false errors, and optimizing clinical practice in breast cancer detection.