EfficientNet-B7 framework for anomaly detection in mammogram images

At present, handling imbalanced data, deciphering complex data patterns, selecting suitable unsupervised learning algorithms, and ensuring computational efficiency are among the challenges associated with identifying anomalies in mammography images. To address these issues, the development of a robust machine learning model with effective training is imperative. This work aims to enhance accuracy and feature extraction mechanisms for the detection of abnormalities in mammogram images. It tackles these challenges and explores potential strategies to mitigate them, ensuring optimal performance of the model. The objective of the system is to employ an efficient technique for anomaly detection in mediolateral oblique (MLO) view and cranial-caudal (CC) view of 46,465 mammography images. The proposed approach introduces ensemble learning, where the ensemble members include the k-nearest neighbors (KNN) and Random Forest models. Additionally, it integrates a deep learning approach utilizing the EfficientNet-B7 architecture through transfer learning. The results of the proposed approach indicate a 18.72% reduction in memory consumption and an impressive accuracy of 98.86%. This outperforms the standalone system, where memory consumption and accuracy stand at 22.84% and 95.81%, respectively.