Decomposition-based multi-modal image fusion for breast cancer classification using AlexNet and MCFO filter

Abstract Breast cancer is one of the most common and deadly cancers, affecting millions worldwide. Early and accurate detection is essential for effective treatment and improved patient outcomes. Advances in medical imaging technologies, such as Digital Mammography (DM), Ultrasound (US), and Magnetic Resonance Imaging (MRI) provide clinicians with detailed information about breast tumors and surrounding tissues. However, merging and analyzing these multimodal images pose challenges. Medical image fusion combines images from different modalities to improve quality, reduce noise and redundancy, and support more precise clinical decisions. In this study, three models were developed to evaluate feature extraction strategies: Model A uses an AlexNet architecture, Model B employs a LeNet-5 architecture, and Model C incorporates a DenseNet-121 architecture. All models are integrated with a decomposition method, such as PCA or DWT, for image fusion into three main categories: normal, benign, or malignant. The Modified Central Forced Optimization (MCFO) filter is employed to enhance diagnostic accuracy. Our framework was tested on a new dataset from Baheya Hospital in Egypt, which includes high-quality, annotated images. Results show that combining DWT-based methods with AlexNet and the MCFO filter achieves top performance, with an accuracy of 97.4%, a precision of 95%, a Recall of 96%, a F1 Score of 93%, and an ROC score of 96.95%, with minimal loss, demonstrating strong generalization and stability across epochs. These findings highlight the superior performance of the DWT-based approach with AlexNet and MCFO compared to other methods.