An approach with Visual and Tabular Mamba to multimodal medical data using Mixed Fusion

The article presents a novel approach for multimodal medical data integration in cancer classification using the Mamba architecture, which incorporates a Mixed Fusion model for enhanced interpretability. Two Mamba variants are utilized: one for visual processing of lesion images and another for tabular data processing that combines clinical information with image-derived probabilities. Experimental results on the PAD-UFES-20 and NDB-UFES datasets show competitive performance, with notable improvements in recall, suggesting that Mamba-based models are effective for sensitive medical classification tasks while also facilitating interpretative methods like SHAP.