人工智能在糖尿病视网膜病变筛查中的应用进展

Application progress of artificial intelligence in the screening of diabetic retinopathy

人工智能(AI)是研究、开发用于模拟、拓展人类智能的理论、方法、技术及应用系统的一门新兴科学技术,目前已在智能医疗方面取得了重大突破,且其在糖尿病视网膜病变(DR)、早产儿视网膜病变等眼底疾病的诊疗中展现出了巨大潜力。国内外已开展多项AI技术应用于DR筛查的临床试验,其不仅具有较高的准确率,而且可为医生节省阅片时间,降低社会、医疗工作及患者的负担。但目前由于缺乏针对DR智能诊断技术的评估系统,AI系统的准确性尚缺乏大数据的验证。其次,大部分研究的彩色眼底照片拍摄范围为后极部45°,仅显示了最容易发病的区域,使得有些病变无法被检测。此外,目前DR筛查系统尚未应用于临床,大多数处于前瞻性研究试验阶段,从研发环境到临床仍存在诸多障碍,医生无法使用真实的患者数据来评估AI系统,因此在临床中并未普及。未来可进一步完善和建立DR筛查的算法以及诊断模型,使DR的AI筛查更加准确。

Artificial intelligence(AI)is an emerging science and technology that studies and develops theories,methods,technologies,and application systems for simulating and expanding human intelligence.AI has made great breakthroughs in the field of intelligent medicine,and has shown great potential in the diagnosis and treatment of diabetic retinopathy(DR),retinopathy of prematurity,and other fundus diseases.A number of clinical trials on the application of AI technologies to DR screening have been carried out in the domestic and overseas,which not only have a high accuracy rate,but also save doctors'reading time and reduce the burden of society,medical work and patients.However,due to the lack of evaluation system for DR intelligent diagnosis technology,the accuracy of AI system still lacks of big data verification.Secondly,most of the color fundus photographs are taken in the posterior 45°,which only show the most vulnerable areas,making some lesions undetectable.In addition,the current DR screening system has not yet been applied to the clinic,most of which are in the stage of prospective research and trials.There are still many obstacles from the environment to the hospital or the clinic.Doctors cannot use real patient data to evaluate the AI system,so it is not popular in clinical practice.In the future,DR screening algorithms and diagnostic models can be further improved and established to make DR AI screening more accurate.