DCC

Content

D20A0003

Bloodless AI-assisted ECG System to Rapidly Detect Life-threatening Dyskalemia

Invention Description
We used the 66,321 ECG records from 40,180 patients in emergency department during May 2011 to December 2016 for conducting the experiment. A deep learning based model, ECG12Net, was developed to predict a value of potassium concentration. ECG12Net presented the mean absolute error of 0.531 with the sensitivities of severe hypokalemia/hyperkalemia of 95.6% and 84.5%, respectively. The specificities of detecting hypokalemia and hyperkalemia were 81.6% and 96.0%, respectively. The balance accuracies of our AI system in a humanmachine competition were 80.4%/82.7% in hypo/hyperkalemia, which was more than 10% better than that of our best clinician (66.7%/70.6%).

Competitive Advantages
1. Traditional laboratory test of potassium concentration takes more than half hour to complete the test. Our ECG based AI system provide a prediction of severe dyskalemias within 1 minute, and the performance is significantly higher than clinicians.
2. A large scale prospective study has been validated the improvement of patient management. The AI-alarm system in our hospital has actively identified 22 severe hypokalemia and 36 severe hyperkalemia patients, and the clinical outcomes were significantly better than those before AIalarm system online.
3. To compare with the worldwide-only product from Mayo Clinic, the sensitivity and specificity of our AI system on hyperkalemia detection is similar with theirs. Important, our AI system is able to detect hypokalemia and provides a value of potassium concentration, which is unique worldwide.


Key Publication
Lin CS#, Lin C#, Fang WH, Hsu CJ, Chen SJ, Huang KH, Lin WS, Tsai CS, Kuo CC, Chau T, Yang SJ, Lin SH* (2020, Mar) A Deep-Learning Algorithm (ECG12Net) for Detecting Hypokalemia and Hyperkalemia by Electrocardiography: Algorithm Development. JMIR Medical Informatics, 8(3):e15931. Intellectual Property: Patent applications pending in U.S.A. and Taiwan.

Business Opportunity
License and/or Sponsored Research

※ If you are interested in our technologies, please contact us here.

※If you want to download the file. Please Login/Register