Taesic Lee1, Juwon Kim2, Young Uh3, Hyunju Lee4. 1. Department of Biomedical Science and Engineering, Gwangju Institute of Science and Technology, Gwangju, South Korea. 2. Department of Laboratory Medicine, Yonsei University Wonju College of Medicine, Wonju, South Korea. 3. Department of Laboratory Medicine, Yonsei University Wonju College of Medicine, Wonju, South Korea.. Electronic address: u931018@yonsei.ac.kr. 4. Department of Biomedical Science and Engineering, Gwangju Institute of Science and Technology, Gwangju, South Korea.; School of Electrical Engineering and Computer Science, Gwangju Institute of Science and Technology, Gwangju, South Korea.. Electronic address: hyunjulee@gist.ac.kr.
Abstract
BACKGROUND: LDL cholesterol (LDL-C) has been mainly estimated using the Friedewald equation, and other equations have recently been developed to complement the Friedewald equation. The present study aims to employ a deep neural network (DNN) to improve LDL-C estimation. METHODS: We used two independent datasets obtained from the Korean National Health and Nutrition Examination Survey and the Wonju Severance Christian Hospital as training and test datasets, respectively. We used the training dataset to construct the DNN architecture, which takes three input values of total cholesterol, HDL cholesterol, and triglyceride, and estimates LDL-C as the output. The model consists of six hidden layers, and each hidden layer has 30 nodes. The performance of the DNN model constructed by the training dataset was measured using the test dataset. RESULTS: In fivefold cross-validation using the training dataset, the DNN model showed the lowest mean and median squared errors compared to the Friedewald equation and Novel method. For the independent test dataset, our DNN model outperformed other existing methods on the basis of mean and median squared errors. CONCLUSIONS: The DNN model provided the most accurate estimation of LDL-C compared to other existing methods including the Friedewald and Novel methods.
BACKGROUND:LDL cholesterol (LDL-C) has been mainly estimated using the Friedewald equation, and other equations have recently been developed to complement the Friedewald equation. The present study aims to employ a deep neural network (DNN) to improve LDL-C estimation. METHODS: We used two independent datasets obtained from the Korean National Health and Nutrition Examination Survey and the Wonju Severance Christian Hospital as training and test datasets, respectively. We used the training dataset to construct the DNN architecture, which takes three input values of total cholesterol, HDL cholesterol, and triglyceride, and estimates LDL-C as the output. The model consists of six hidden layers, and each hidden layer has 30 nodes. The performance of the DNN model constructed by the training dataset was measured using the test dataset. RESULTS: In fivefold cross-validation using the training dataset, the DNN model showed the lowest mean and median squared errors compared to the Friedewald equation and Novel method. For the independent test dataset, our DNN model outperformed other existing methods on the basis of mean and median squared errors. CONCLUSIONS: The DNN model provided the most accurate estimation of LDL-C compared to other existing methods including the Friedewald and Novel methods.