Linking Electrocardiogram and Echocardiogram: Comparing Classical Machine Learning and Deep Learning Neural Networks for the Detection of Regional Wall Motion Abnormalities
Historically, electrocardiogram (ECG) datasets have been created based on physicians’ interpretation of the ECG, which may introduce human biases and errors. Nightingale Open Science provides an open-source ECG dataset linking to an imaging marker, regional wall motion abnormality (RWMA),...
Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11096598/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849240269610811392 |
|---|---|
| author | Shantanu M. Joshi Hana R. Shaik Shivam Rai Sharma Philip Strong Uma Srivatsa Imo Ebong Hyoyoung Jeong Chen-Nee Chuah Lihong Mo |
| author_facet | Shantanu M. Joshi Hana R. Shaik Shivam Rai Sharma Philip Strong Uma Srivatsa Imo Ebong Hyoyoung Jeong Chen-Nee Chuah Lihong Mo |
| author_sort | Shantanu M. Joshi |
| collection | DOAJ |
| description | Historically, electrocardiogram (ECG) datasets have been created based on physicians’ interpretation of the ECG, which may introduce human biases and errors. Nightingale Open Science provides an open-source ECG dataset linking to an imaging marker, regional wall motion abnormality (RWMA), that is primarily associated with myocardial ischemia or infarction. RWMA refers to an area of the heart muscle that does not contract properly and can be identified on an imaging modality such as an echocardiogram or cardiac magnetic resonance imaging (MRI). Our study aimed to predict RWMA using both classical machine learning (ML) methods and a one-dimensional convolutional neural network (1D CNN) model on the 12-lead ECG data from 3,750 unique patients provided by the Nightingale Open Science platform. Of the 3,750 patients, 341 (9.1%) were diagnosed with positive RWMA on their echocardiogram within a year of the ECG. After filtering out patients with missing data, we identified 313 patients with positive RWMA label and randomly undersampled 313 from the remaining patients to keep the data balanced. We repeated the experiments with five random selections of the 313 patients without RWMA to ensure the robustness of the models. Our results indicate that the optimized 1D CNN model achieved an area under the receiver operating characteristic curve (AUROC) of 73.40% (1D CNN) versus 63.58% (Random Forest), an area under the precision-recall curve (AUPRC) of 72.45% (1D CNN) versus 72.64% (Random Forest), and accuracy of 71.15% (1D CNN) versus 63.58% (Random Forest). The Random Forest model revealed that QT duration and maximal amplitude differential of ST elevation or depression as the key features associated with RWMA. This study demonstrated that CNN models and classical ML methods offer complementary advantages. Additionally, we compared the performance, training time and model complexity of 1D CNN model with the latest foundation model, HuBERT-ECG. This is a preliminary step for further elucidation of the relationship between ECG and echocardiogram. Future studies are needed to determine the optimal CNN architecture that balances the target classifier performance and computational efficiency for ease of deployment with larger datasets. |
| format | Article |
| id | doaj-art-018c9a24567242aab4a402c1810f1a9b |
| institution | Kabale University |
| issn | 2169-3536 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj-art-018c9a24567242aab4a402c1810f1a9b2025-08-20T04:00:40ZengIEEEIEEE Access2169-35362025-01-011313451913452810.1109/ACCESS.2025.359269311096598Linking Electrocardiogram and Echocardiogram: Comparing Classical Machine Learning and Deep Learning Neural Networks for the Detection of Regional Wall Motion AbnormalitiesShantanu M. Joshi0https://orcid.org/0009-0001-0110-6515Hana R. Shaik1https://orcid.org/0009-0008-6572-4882Shivam Rai Sharma2https://orcid.org/0009-0003-8550-8443Philip Strong3https://orcid.org/0009-0007-1982-9475Uma Srivatsa4https://orcid.org/0000-0003-0378-2346Imo Ebong5https://orcid.org/0000-0001-9687-7541Hyoyoung Jeong6https://orcid.org/0000-0002-1808-7824Chen-Nee Chuah7https://orcid.org/0000-0002-2772-387XLihong Mo8https://orcid.org/0000-0001-6701-0594Department of Computer Science, University of California at Davis, Davis, CA, USADepartment of Electrical and Computer Engineering, University of California at Davis, Davis, CA, USADepartment of Computer Science, University of California at Davis, Davis, CA, USADepartment of Internal Medicine, Division of Cardiovascular Medicine, University of California, Davis, Sacramento, CA, USADepartment of Internal Medicine, Division of Cardiovascular Medicine, University of California, Davis, Sacramento, CA, USADepartment of Internal Medicine, Division of Cardiovascular Medicine, University of California, Davis, Sacramento, CA, USADepartment of Electrical and Computer Engineering, University of California at Davis, Davis, CA, USADepartment of Electrical and Computer Engineering, University of California at Davis, Davis, CA, USADepartment of Obstetrics and Gynecology, University of California at Davis, Davis, CA, USAHistorically, electrocardiogram (ECG) datasets have been created based on physicians’ interpretation of the ECG, which may introduce human biases and errors. Nightingale Open Science provides an open-source ECG dataset linking to an imaging marker, regional wall motion abnormality (RWMA), that is primarily associated with myocardial ischemia or infarction. RWMA refers to an area of the heart muscle that does not contract properly and can be identified on an imaging modality such as an echocardiogram or cardiac magnetic resonance imaging (MRI). Our study aimed to predict RWMA using both classical machine learning (ML) methods and a one-dimensional convolutional neural network (1D CNN) model on the 12-lead ECG data from 3,750 unique patients provided by the Nightingale Open Science platform. Of the 3,750 patients, 341 (9.1%) were diagnosed with positive RWMA on their echocardiogram within a year of the ECG. After filtering out patients with missing data, we identified 313 patients with positive RWMA label and randomly undersampled 313 from the remaining patients to keep the data balanced. We repeated the experiments with five random selections of the 313 patients without RWMA to ensure the robustness of the models. Our results indicate that the optimized 1D CNN model achieved an area under the receiver operating characteristic curve (AUROC) of 73.40% (1D CNN) versus 63.58% (Random Forest), an area under the precision-recall curve (AUPRC) of 72.45% (1D CNN) versus 72.64% (Random Forest), and accuracy of 71.15% (1D CNN) versus 63.58% (Random Forest). The Random Forest model revealed that QT duration and maximal amplitude differential of ST elevation or depression as the key features associated with RWMA. This study demonstrated that CNN models and classical ML methods offer complementary advantages. Additionally, we compared the performance, training time and model complexity of 1D CNN model with the latest foundation model, HuBERT-ECG. This is a preliminary step for further elucidation of the relationship between ECG and echocardiogram. Future studies are needed to determine the optimal CNN architecture that balances the target classifier performance and computational efficiency for ease of deployment with larger datasets.https://ieeexplore.ieee.org/document/11096598/ClassificationCNNdeep learningechocardiogramelectrocardiogramECG |
| spellingShingle | Shantanu M. Joshi Hana R. Shaik Shivam Rai Sharma Philip Strong Uma Srivatsa Imo Ebong Hyoyoung Jeong Chen-Nee Chuah Lihong Mo Linking Electrocardiogram and Echocardiogram: Comparing Classical Machine Learning and Deep Learning Neural Networks for the Detection of Regional Wall Motion Abnormalities IEEE Access Classification CNN deep learning echocardiogram electrocardiogram ECG |
| title | Linking Electrocardiogram and Echocardiogram: Comparing Classical Machine Learning and Deep Learning Neural Networks for the Detection of Regional Wall Motion Abnormalities |
| title_full | Linking Electrocardiogram and Echocardiogram: Comparing Classical Machine Learning and Deep Learning Neural Networks for the Detection of Regional Wall Motion Abnormalities |
| title_fullStr | Linking Electrocardiogram and Echocardiogram: Comparing Classical Machine Learning and Deep Learning Neural Networks for the Detection of Regional Wall Motion Abnormalities |
| title_full_unstemmed | Linking Electrocardiogram and Echocardiogram: Comparing Classical Machine Learning and Deep Learning Neural Networks for the Detection of Regional Wall Motion Abnormalities |
| title_short | Linking Electrocardiogram and Echocardiogram: Comparing Classical Machine Learning and Deep Learning Neural Networks for the Detection of Regional Wall Motion Abnormalities |
| title_sort | linking electrocardiogram and echocardiogram comparing classical machine learning and deep learning neural networks for the detection of regional wall motion abnormalities |
| topic | Classification CNN deep learning echocardiogram electrocardiogram ECG |
| url | https://ieeexplore.ieee.org/document/11096598/ |
| work_keys_str_mv | AT shantanumjoshi linkingelectrocardiogramandechocardiogramcomparingclassicalmachinelearninganddeeplearningneuralnetworksforthedetectionofregionalwallmotionabnormalities AT hanarshaik linkingelectrocardiogramandechocardiogramcomparingclassicalmachinelearninganddeeplearningneuralnetworksforthedetectionofregionalwallmotionabnormalities AT shivamraisharma linkingelectrocardiogramandechocardiogramcomparingclassicalmachinelearninganddeeplearningneuralnetworksforthedetectionofregionalwallmotionabnormalities AT philipstrong linkingelectrocardiogramandechocardiogramcomparingclassicalmachinelearninganddeeplearningneuralnetworksforthedetectionofregionalwallmotionabnormalities AT umasrivatsa linkingelectrocardiogramandechocardiogramcomparingclassicalmachinelearninganddeeplearningneuralnetworksforthedetectionofregionalwallmotionabnormalities AT imoebong linkingelectrocardiogramandechocardiogramcomparingclassicalmachinelearninganddeeplearningneuralnetworksforthedetectionofregionalwallmotionabnormalities AT hyoyoungjeong linkingelectrocardiogramandechocardiogramcomparingclassicalmachinelearninganddeeplearningneuralnetworksforthedetectionofregionalwallmotionabnormalities AT chenneechuah linkingelectrocardiogramandechocardiogramcomparingclassicalmachinelearninganddeeplearningneuralnetworksforthedetectionofregionalwallmotionabnormalities AT lihongmo linkingelectrocardiogramandechocardiogramcomparingclassicalmachinelearninganddeeplearningneuralnetworksforthedetectionofregionalwallmotionabnormalities |