Scholarlink

<p>In this paper, a new hybrid fusion model has been proposed to achieve early chronic disease detection, i.e. diabetes, cardiovascular diseases and cancer, through the usage of multi-modal health data. The model, trained on a combination of electronic health records (MIMIC-III), medical imaging (CheXpert), wearable sensor time-series (Synthetic Wearable), and genomic profiles (TCGA), uses convolutional neural networks, long short-term memory networks, and transformers to make robust predictions. On test sets, the model obtains AUC-ROC values of 0.89–0.92, accuracies of 0.88–0.89, and F1-scores of 0.871–0.881 and outperforms single-modal baselines (CNN, LSTM, logistic regression). Confusion matrices indicate high true positive rates, which guarantee trustful detection. The tools of interpretability, such as SHAP, identify such important features as fasting glucose, cardiomegaly, and BRCA1 mutations, which is consistent with clinical expectations. The practical uses of the case studies include directing early interventions on patients at risk. This method provides a framework of scalable and interpretable precision medicine, which has potentials to improve clinical decision-making and patient outcomes due to timely management of chronic diseases.</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

Deep neural networks for early detection of chronic diseases using multi-modal health data

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Dharmendra Kumar

摘要

In this paper, a new hybrid fusion model has been proposed to achieve early chronic disease detection, i.e. diabetes, cardiovascular diseases and cancer, through the usage of multi-modal health data. The model, trained on a combination of electronic health records (MIMIC-III), medical imaging (CheXpert), wearable sensor time-series (Synthetic Wearable), and genomic profiles (TCGA), uses convolutional neural networks, long short-term memory networks, and transformers to make robust predictions. On test sets, the model obtains AUC-ROC values of 0.89–0.92, accuracies of 0.88–0.89, and F1-scores of 0.871–0.881 and outperforms single-modal baselines (CNN, LSTM, logistic regression). Confusion matrices indicate high true positive rates, which guarantee trustful detection. The tools of interpretability, such as SHAP, identify such important features as fasting glucose, cardiomegaly, and BRCA1 mutations, which is consistent with clinical expectations. The practical uses of the case studies include directing early interventions on patients at risk. This method provides a framework of scalable and interpretable precision medicine, which has potentials to improve clinical decision-making and patient outcomes due to timely management of chronic diseases.