Deep meta-learning framework with U-Net segmentation and optimization-based feature selection for breast cancer detection in MRI images.

Background: Breast cancer is one of the most prevalent cancers worldwide and continues to pose significant challenges to global healthcare systems. The survival rate of patients largely depends on early detection, accurate diagnosis, and timely treatment. Medical imaging techniques such as mammography, ultrasound, and magnetic resonance imaging (MRI) are widely used for detecting abnormalities in breast tissue. However, many existing breast cancer detection approaches suffer from limitations such as high computational complexity, long tumor detection time, inaccurate predictions, and high false-positive rates. These challenges highlight the need for more reliable and efficient automated diagnostic systems. The aim of this study is to develop an efficient deep meta-learning framework for accurate breast cancer detection and classification using MRI images. The proposed system integrates preprocessing, tumor segmentation, feature extraction, optimization, and deep learning models to improve diagnostic accuracy, reduce false-positive rates, and support early clinical decision-making in computer-aided diagnosis systems. Methods: This work proposes a deep meta-learning framework for precise breast cancer detection and classification using MRI images. The framework integrates multiple stages to improve diagnostic accuracy. Initially, Adaptive Median Filtering (AMF) is applied to enhance image quality and remove noise during preprocessing. Subsequently, U-Net-based segmentation is used to accurately identify tumor regions. Growth Distribution Depth (GDD) is employed to extract discriminative features from the segmented regions, and Salp Swarm Optimization (SSO) is used to select the most relevant features. Finally, deep meta-learning-based classification is performed using pre-trained convolutional neural network (CNN) models, including AlexNet, VGG16, and LeNet, which are fine-tuned on breast cancer datasets to improve classification performance. Results: Experimental results demonstrate that the proposed framework achieves superior performance compared to existing methods. Among the evaluated models, the VGG16-based classifier shows the best performance, achieving an accuracy of 98.82%, precision of 97.1%, recall of 99.11%, and an F1-score of 98.25%. The integration of tumor-focused segmentation and optimization-based feature selection significantly improves classification accuracy and reduces false detections. Conclusions: The proposed deep meta-learning framework provides an effective and reliable approach for automated breast cancer detection from MRI images. By combining advanced preprocessing, segmentation, feature extraction, optimization, and deep learning classification techniques, the framework enhances diagnostic accuracy and efficiency. The results suggest that the proposed method has strong potential to support computer-aided diagnosis systems and assist clinicians in early breast cancer detection.