Externally Validated Deep Learning Model to Identify Prodromal Parkinson’s Disease from Electrocardiogram
- Python 3.7.7
- scikit-learn 0.22.1
- scipy 1.5.2
- tensorflow 1.15.0
- tensorflow-base 1.15.0
- tensorflow-estimator 1.15.1
- keras 2.2.4
- keras-applications 1.0.8
- keras-base 2.2.4
Using only a simple 10-second ECG we built a predictive model that correctly classified individuals with prodromal PD with moderate accuracy. The model was effective in an independent cohort, particularly closer to disease diagnosis. Standard ECGs may help to identify individuals with prodromal PD for inclusion in disease-modifying therapeutic trials.
- Required Python versions and libraries are listed above.
- Get the code $ git clone the repo and install the dependencies
- For the predicted risk scores and labels, execute the code below in the local repo directory
- Please make sure that your input ECG follows the format provided here. e.g. 12-Lead, 10 seconds, at 500Hz. following the lead order of: I, II, III, aVR, aVL, aVF, V1, V2, V3, V4, V5, V6
python Github-ECG-AI_PD_Risk_Prediction.py sample_ECG.npy
The py file would provide risk scores and labels to your local directory as csv files.
If you find this code useful, please cite the following paper:
Externally Validated Deep Learning Model to Identify Prodromal Parkinson’s Disease from ECG
Ibrahim Karabayir, Fatma Gunturkun, Samuel M Goldman*, Rishikesan Kamaleswaran, Robert L Davis, Kalea Colletta, Lokesh Chintala, John L Jefferies, Kathleen Bobay, G. Webbster Ross, Helen Petrovitch, Kamal Masaki, Caroline M. Tanner, Oguz Akbilgic*
*Corresponding Authors
For any feedback, or bug report, please don't hesitate to contact with the author, Dr. Ibrahim Karabayir