Enhanced character-level deep convolutional neural networks for cardiovascular disease prediction

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Machine Learning

Background:

• Electronic medical records (EMRs) contain valuable patient data for disease prediction.
• Cardiovascular disease (CVD) is a global health concern, necessitating efficient diagnostic tools.
• Automated clinical text processing can enhance the accuracy of CVD diagnosis.

Purpose of the Study:

• To develop an advanced model for predicting cardiovascular disease (CVD) using electronic medical records (EMRs).
• To improve the accuracy and efficiency of CVD diagnosis through automated risk factor extraction and prediction.

Main Methods:

• Proposes an Enhanced Character-level Deep Convolutional Neural Networks (EnDCNN) model.
• Utilizes text region embedding for mapping risk factors and labels into vectors.
• Incorporates downsampling and shortcut connections with pre-activation for improved training efficiency.

Main Results:

• Achieved an F-score of 0.9073 for the risk factor identification and extraction model.
• The EnDCNN prediction model achieved an F-score of 0.9516.
• Demonstrated superior performance compared to previous CVD prediction methods.

Conclusions:

• Character-level models with text region embedding effectively represent risk factors.
• Downsampling is crucial for enhancing the training efficiency of deep convolutional neural networks (CNNs).
• The proposed model architecture facilitates efficient training without dimension-matching issues.