Related Experiment Video
Updated: Jun 29, 2026
09:17
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
22.2K
Interpreting Deep Neural Networks in Diabetic Retinopathy Grading: A Comparison with Human Decision Criteria
Sangeeta Biswas1, Md Ahanaf Arif Khan1, Md Hasnain Ali1
1Faculty of Engineering, University of Rajshahi, Rajshahi 6205, Bangladesh.
Life (Basel, Switzerland)
|September 27, 2025
Summary
This study investigates if deep neural network-based Automatic Diabetic Retinopathy Classifiers (ADRCs) use the same criteria as human experts. Using explainable AI, we analyzed ADRC decisions on fundus images to ensure reliable DR grading.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Diabetic retinopathy (DR) is a leading cause of vision loss globally.
- Automatic Diabetic Retinopathy Classifiers (ADRCs) using deep neural networks (DNNs) show promise for widespread screening.
- The reliability of DNN-based ADRCs is questioned regarding their alignment with human retina professionals' (HRPs) grading criteria.
Purpose of the Study:
- To investigate whether DNN-based ADRCs utilize the same diagnostic criteria as HRPs for DR severity grading.
- To assess the reliability of ADRCs by comparing their decision-making processes with human expert evaluations.
Main Methods:
- Experimentation on publicly available fundus image datasets.
- Development and utilization of MobileNet-based ADRCs.
- Application of two eXplainable artificial intelligence (XAI) techniques: Gradient-weighted Class Activation Map (Grad-CAM) and Integrated Gradients (IG) to analyze ADRC outputs.
Main Results:
- Analysis revealed discrepancies in the criteria used by DNN-based ADRCs compared to HRPs.
- XAI techniques provided insights into the visual features influencing ADRC decisions.
- The study highlights potential issues in the clinical reliability of current ADRCs.
Conclusions:
- DNN-based ADRCs may not consistently employ the same diagnostic criteria as human experts in DR grading.
- Further research and validation are needed to ensure ADRCs align with clinical standards for reliable DR screening.
- XAI methods are crucial for understanding and improving the transparency of AI in medical diagnostics.
Related Concept Videos
01:26Visual System
Light enters the eye through the cornea, a transparent, dome-shaped surface covering the surface of the eyeball that helps to direct and focus incoming light. This light is then channeled toward the pupil, an adjustable opening whose size is controlled by the iris. The iris, a pigmented muscle, regulates the amount of light entering the eye by contracting or dilating the pupil, thereby ensuring optimal light levels for clear vision.
Once through the pupil, the light passes through the lens, a...
Once through the pupil, the light passes through the lens, a...
01:27Diabetic Retinopathy
DefinitionDiabetic retinopathy is a microvascular complication of diabetes affecting the retinal blood vessels.Risk FactorsDiabetic retinopathy is present in almost all individuals with type 1 diabetes and more than 60% of those with type 2 diabetes after two decades of disease.The risk increases with poor glycemic control, hypertension, dyslipidemia, smoking, pregnancy, and puberty.Although cataracts and glaucoma are also more frequent in people with diabetes, retinopathy remains the leading...
01:22Diabetic Neuropathy
DefinitionDiabetic neuropathy is nerve damage caused by long-standing diabetes mellitus. It results directly from prolonged high blood sugar levels.PathophysiologyThe pathophysiology of diabetic neuropathy involves both metabolic and vascular disturbances triggered by chronic hyperglycemia.Metabolic injury: Elevated glucose levels activate the polyol pathway within nerve cells, leading to the accumulation of sorbitol and fructose. This increases oxidative stress, disrupts normal nerve...