Artificial intelligence-based classification of cardiac autonomic neuropathy from retinal fundus images in patients with diabetes: The Silesia Diabetes Heart Study
Authors' conclusion
AI-based algorithms utilising retinal images can differentiate with high accuracy patients with CAN. AI analysis of fundus images to detect CAN may be implemented in routine clinical practice to identify patients at the highest CV risk.
Study context
- Population
- Patients with diabetes mellitus (DM), Silesia Diabetes-Heart Project; Cardiac autonomic neuropathy (CAN) diagnosed by standard cardiovascular (CV) autonomic reflex tests
- Participants
- 229 patients2,275 retinal images
- Comparison
- Artificial intelligence (AI) deep learning on retinal fundus images vs CV autonomic reflex tests for CAN
- Outcome
- Binary classification of CAN
Methodological detailssingle-centre, observational, deep learning, Multiple Instance Learning, unseen test set
- Single-centre
- Observational
- Deep learning
- Multiple Instance Learning
- Unseen test set
Performance
As reported in the abstract
Binary classification of CAN (test set)
| Test or model | CAN cases correctly identified | non-CAN cases correctly identified | AUC |
|---|---|---|---|
| ResNet 18 | 93% | 89% | 0.87 (95% CI 0.74-0.97) |
ResNet 18
CAN cases correctly identified93%
non-CAN cases correctly identified89%
AUC0.87 (95% CI 0.74-0.97)
Further findings
| Outcome | ResNet 18 | ResWide 50 |
|---|---|---|
| Classification of definite or severe CAN (dsCAN) | ||
| dsCAN cases correctly classified | 78% | — not stated in the abstract |
| Cases without dsCAN correctly classified | 93% | — not stated in the abstract |
| AUC | 0.94 (95% CI 0.86-1.00) | 0.91 (95% CI 0.73-1.00) |
| Sensitivity for dsCAN | — not stated in the abstract | 89% |
— not stated in the abstract
Original publication
Artificial intelligence-based classification of cardiac autonomic neuropathy from retinal fundus images in patients with diabetes: The Silesia Diabetes Heart Study. Cardiovasc Diabetol. Published 10 August 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →