SOLD-MMD: Wolf Pack Search Optimized Deep Learning Framework for Brain Disease Detection Using Multi-modal Data

Brain disease detection remains a complex challenge due to the heterogeneous nature of multimodal healthcare data, including medical images, electronic health records (EHR), and physiological signals. The effective integration of these diverse modalities is crucial to accurate and timely clinical decisions. Despite this, existing models in this domain have key shortcomings, such as poor data fusion strategies, redundant data storage, and insufficient exploitation of temporal and spatial dependencies. To address these drawbacks, a novel wolf pack Search Optimized deep Learning model for brain disease Detection using a Multi-Modal Data (SOLD-MMD) framework is proposed. The proposed approach creates a Merged Multi-Modal Set (M3Set) by combining clinical image, EHR, and signal data from the MIMIC database. A Rough Set-based Heterogeneous Method is used to detect and eliminate duplicate patient records by ensuring data consistency and reducing computational overhead. A hybrid Spiking Convolutional Neural Network-based Bidirectional Long Short-Term Memory (SCNN-BiLSTM) model is used to extract both spatial and sequential features. The hyperparameters of the SCNN-BiLSTM model are tuned using the Wolf Pack Search Optimization (WPSO) algorithm. SOLD-MMD is evaluated with several metrics, including accuracy, precision, and recall. SOLD-MMD achieves outstanding results across all evaluation metrics, with 98.59% Accuracy, 97.18% Specificity, 96.96% Precision, 97.41% Recall, and 96.10% F1 Score.