<p>Multimodal breast cancer refers to the classification of breast cancer by combining various imaging modalities whose results improve the sensitivity of the diagnosis. It enhances early identification and categorization using complementary features from multiple sources. This approach involves precise alignment of heterogeneous data and introduces computational complexity. To address these obstacles, this paper suggests a new Fractional-Order Three-Triangle Multi-delayed Neural Networks with Hunger Games Search Optimization (FOT2MNN-HGSO) to promote multimodal breast cancer classification accuracy. First, mammogram and Ultrasound images are gathered from MIAS and UDIAT databases. Then, Structure-Aware Adaptive Bilateral Texture Filtering (SA2BTF) is applied for pre-processing for image quality enhancement and noise reduction. Subsequently, Geometric Algebra Transformer (GAT) is employed for segmentation to precisely outline regions of interest through the recording of geometric and spatial relationships. Next, the Heterogeneous Edge-Enhanced Graph Attention Network (HE2GAN) is employed for feature extraction to record intricate node and edge relations between imaging modalities. Next, Fractional-Order Three-Triangle Multi-delayed Neural Networks (FOT2MNN) is employed for classification in order to enhance differentiation between malignant and benign cancers. At last, Hunger Games Search Optimization (HGSO) is employed to optimize the parameters of the model and enhance the classification performance. The approach resulted in an F1-score value of 98.9%, an error rate of 0.1%, and an accuracy level of 99.9%. The MIAS and UDIAT datasets demonstrate a significant improvement in diagnostic reliability and a reduction in false positives. For multimodal breast cancer classification, the FOT2MNN-HGSO model provides a highly accurate, noise-resistant, and reliable solution with great potential for clinical decision-making support and computer-assisted medical diagnosis.</p>