Deep ensemble transfer learning-based framework for mammographic image classification

This research intends to provide a method for clinical decision support systems that can accurately classify benign and malignant mass from breast X-ray images. The model was initially trained and assessed using distinct convolutional neural network (CNN) models. Based on the comparative analysis, the best models were then selected and used for further implementation. The work employs an average, weighted average and concatenation strategy that seeks to merge pre-trained CNN, relying on the transfer learning technique to construct a highly accurate ensemble model. The models make it possible to save and use the information gained from a pre-trained CNN for a new task, namely breast mammogram classification. A benchmark datasets such as MIAS, CBIS-DDSM and a private dataset with two classes, benign and malignant, were used in the proposed approach. To assess the proposed model's efficacy, several generic assessment techniques were employed. Our ensemble model outperforms other state-of-the-art methods with an overall accuracy (99.76%), Sensitivity (99.84%), Specificity(99.76%), Precision(99.76%) and F1-Score(99.76%).