This review presents a comprehensive analysis of deep learning-based multimodal fusion techniques for medical image classification. The paper explores three main fusion schemes: input fusion, intermediate fusion (including single-level, hierarchical, and attention-based fusion), and output fusion. It evaluates the performance of these techniques to determine their suitability for various multimodal fusion scenarios and application domains. The review also discusses challenges related to network architecture selection, handling incomplete multimodal data, and the potential limitations of multimodal fusion. It highlights the promising future of Transformer-based multimodal fusion techniques and provides recommendations for future research in this rapidly evolving field. The paper discusses the complementary relationships among prevalent clinical modalities and outlines the three main fusion schemes for multimodal classification networks. It also reviews the development trends of multimodal medical image classification, noting the increasing number of publications and the growing focus on brain-related studies. The paper presents a detailed taxonomy of multimodal information fusion, categorizing it into input fusion, intermediate fusion, and output fusion. It further segments intermediate fusion into single-level, hierarchical, and attention-based fusion. The paper summarizes the five strategies of multimodal fusion: input fusion, single-level fusion, hierarchical fusion, attention-based fusion, and output fusion. These fusion methods can be applied to any multimodal classification problem in medicine, allowing for greater flexibility and potential for improved results. The paper also presents the prevailing challenges and predicts future trends in multimodal fusion. It discusses the importance of pre-processing, information fusion, deep learning backbones, and final classifiers in multimodal classification. The paper also presents evaluation metrics for multimodal fusion tasks, including accuracy, sensitivity, specificity, precision, F1 score, AUC, and Kappa. The paper reviews various multimodal image datasets and their applications in medical image classification. It discusses the characteristics of different imaging modalities and their use in multimodal fusion. The paper also discusses the challenges of multimodal fusion, including the need for accurate registration, the impact of different modalities on classification results, and the limitations of current fusion techniques. The paper highlights the potential of Transformer-based multimodal fusion techniques and provides recommendations for future research in this field.This review presents a comprehensive analysis of deep learning-based multimodal fusion techniques for medical image classification. The paper explores three main fusion schemes: input fusion, intermediate fusion (including single-level, hierarchical, and attention-based fusion), and output fusion. It evaluates the performance of these techniques to determine their suitability for various multimodal fusion scenarios and application domains. The review also discusses challenges related to network architecture selection, handling incomplete multimodal data, and the potential limitations of multimodal fusion. It highlights the promising future of Transformer-based multimodal fusion techniques and provides recommendations for future research in this rapidly evolving field. The paper discusses the complementary relationships among prevalent clinical modalities and outlines the three main fusion schemes for multimodal classification networks. It also reviews the development trends of multimodal medical image classification, noting the increasing number of publications and the growing focus on brain-related studies. The paper presents a detailed taxonomy of multimodal information fusion, categorizing it into input fusion, intermediate fusion, and output fusion. It further segments intermediate fusion into single-level, hierarchical, and attention-based fusion. The paper summarizes the five strategies of multimodal fusion: input fusion, single-level fusion, hierarchical fusion, attention-based fusion, and output fusion. These fusion methods can be applied to any multimodal classification problem in medicine, allowing for greater flexibility and potential for improved results. The paper also presents the prevailing challenges and predicts future trends in multimodal fusion. It discusses the importance of pre-processing, information fusion, deep learning backbones, and final classifiers in multimodal classification. The paper also presents evaluation metrics for multimodal fusion tasks, including accuracy, sensitivity, specificity, precision, F1 score, AUC, and Kappa. The paper reviews various multimodal image datasets and their applications in medical image classification. It discusses the characteristics of different imaging modalities and their use in multimodal fusion. The paper also discusses the challenges of multimodal fusion, including the need for accurate registration, the impact of different modalities on classification results, and the limitations of current fusion techniques. The paper highlights the potential of Transformer-based multimodal fusion techniques and provides recommendations for future research in this field.