Accuracy of an AI Model for Diabetic Retinopathy Screening in Real-life
试验速览
- 阶段
- 不适用
- 状态
- 招募中
- 发起方
- 入组人数
- 100
- 试验地点
- 1
- 主要终点
- PPV
研究概览
简要总结
The increasing prevalence of diabetes mellitus represents a major health problem, especially since around 40% of diabetic patients develop diabetic retinopathy, which severely impairs vision and can lead to blindness. This development could be prevented by annual check-ups and timely referral for treatment. However, there are major differences in the quality of examinations and bottlenecks in examination appointments. A solution to the problem could be the use of artificial intelligence (AI), especially deep learning. Initial studies have shown that deep learning algorithms can be used successfully to detect diabetic retinopathy. However, it remains to be clarified whether the use of AI can achieve a sufficiently high level of accuracy in the detection of retinopathies. Therefore, in the present study, the positive predictive value (PPV), the negative predictive value (NPV), the sensitivity (SEN) and the specificity (SPEZ) of the AI algorithm 'MONA-DR-Model' in the detection of diabetic retinopathy should be measured. In addition, it is to be examined how well the classification into mild and severe retinopathy corresponds and how well this new examination method is accepted by the patients.
详细描述
As part of the study, a 45-degree fundus image is taken for each eye and patient using the 'Crystalvue NFC 600'. The fundus photographs are then analyzed using the 'MONA-DR-Mode'l and classified as "diabetic retinopathy according to AI present (K+)" or "diabetic retinopathy according to AI absent (K-)". These classifications are compared with the results ("diabetic retinopathy according to the doctor present (A+)" or "diabetic retinopathy according to the doctor absent (A-)") of the examinations routinely provided for in the Disease Management Program (DMP) diabetes mellitus type 2 by resident ophthalmologists who work in the period 6 months before and after the fundus photography in the West German Centre of Diabetes and Health (WDGZ) were compared. All patients with the assessment "diabetic retinopathy according to AI present (K+)" or discrepancies with the ophthalmological DMP examination in the outpatient environment are offered a routine appointment at the Marienhospital. There, an eye examination is then carried out by an ophthalmologist and, without knowledge of the previous findings, a reassessment and classification as "diabetic retinopathy according to the doctor present (A+)" or "diabetic retinopathy according to the doctor absent (A-)" is carried out by the AI.
研究设计
- 研究类型
- Observational
- 观察模型
- Case Control
- 时间视角
- Other
入排标准
- 年龄范围
- 18 Years 至 —（Adult, Older Adult）
- 性别
- All
- 接受健康志愿者
- 否
入选标准
- •Diagnosis of diabetes mellitus
- •Diabetes duration ≥ 5 years
- •Age > 18 years old
- •Patient is able to give informed consent
- •Fluent in written and spoken German, or interpreter present
排除标准
- •History of laser treatment
- •Contraindication to the fundus imaging systems used in the study
研究组 & 干预措施
K+A+
diabetic retinopathy according to AI present (K+) AND diabetic retinopathy according to the doctor present (A+)
干预措施: artificial intelligence (AI) algorithm of the MONA DR model (Diagnostic Test)
K+A-
diabetic retinopathy according to AI present (K+) AND diabetic retinopathy according to the doctor absent (A-)
干预措施: artificial intelligence (AI) algorithm of the MONA DR model (Diagnostic Test)
K-A+
diabetic retinopathy according to AI absent (K-) AND diabetic retinopathy according to the doctor present (A+)
干预措施: artificial intelligence (AI) algorithm of the MONA DR model (Diagnostic Test)
K-A-
diabetic retinopathy according to AI absent (K-) AND diabetic retinopathy according to the doctor absent (A-)
干预措施: artificial intelligence (AI) algorithm of the MONA DR model (Diagnostic Test)
结局指标
主要结局
PPV
时间窗: 12 months
positive predictive value
SEN
时间窗: 12 months
sensitivity
NPV
时间窗: 12 months
negative predictive value
SPEZ
时间窗: 12 months
specificity
次要结局
未报告次要终点