IGRNet: A Deep Learning Model for Non-Invasive, Real-Time Diagnosis of Prediabetes through Electrocardiograms
Problem
Prediabetes has mild, easily overlooked symptoms and can progress to diabetes without early intervention, so non-invasive early detection is needed.
Why it matters
Early prediabetes detection enables intervention that may prevent progression to diabetes.
What the AI did
IGRNet, a deep learning model, diagnoses prediabetes non-invasively in real time from a 5-second 12-lead ECG, outperforming AlexNet, GoogLeNet and three traditional machine learning algorithms.
Results
Accuracy 0.781 and AUC 0.777 on independent mixed test set; 0.856 accuracy and 0.825 AUC on normal-weight-range test set.
Impact
Peer-reviewed study with independent test-set validation, but no evidence of clinical adoption.
Record
| Venue | Sensors |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Liyang Wang, Yao Mu, Jing Zhao, Xiaoya Wang, Huilian Che |
| Citations | 40 (OpenAlex, at import time) |
| References | 40 |
| Model or method | IGRNet (deep neural network) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3390/s20092556 |
| Open access | gold |
| Keywords | prediabetes, ECG, deep learning, diagnosis, IGRNet, non-invasive |
Importance rating
significant — thousands to millions affected
Prediabetes progresses to diabetes without early intervention; non-invasive real-time detection could enable prevention at population scale, though abstract gives no prevalence numbers.