Feature fusion based machine learning pipeline to improve breast cancer prediction

Abstract

Early detection of malignant breast cancer can significantly improve the survival chances of the involved patients. Analysis of a non-invasive and non-radioactive modality like ultrasound imaging with the help of Machine Learning(ML) and Artificial Intelligence(AI) techniques can be crucial for achieving such effective early-stage detection of the disease. A feature fusion based approach is proposed in this work, in conjunction with an ML pipeline that systematically deals with various problems like high dimensionality, class imbalance, and hyperparameter tuning, so that efficient benign vs. malignant classification can be performed. Experimental evaluation on two publicly available datasets reveals that the proposed approach is able to outperform state-of-the-art techniques on the classification task with an overall performance of above 95% for all the evaluation metrics under consideration and an AUC of ? 0.99. More specifically, an overall improvement of (1-4)%, (2-10)% and (2-7)% over the current state-of-the-art approaches could be obtained for the Accuracy, AUC and Sensitivity metrics respectively, on both the datasets. Such an efficient approach can provide the necessary real-time decision support to the involved radiologists, making better cancer patient care possible. © 2022, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.