Scholarlink

<p>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.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

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

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月12日

查看全文
  • S. Devi,
  • Ahilan Appathurai,
  • M. Ramya Devi,
  • Narayanaperumal Muthukumaran

摘要

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.