Credit: Unsplash/Perchek Industrie
Practical applications of artificial intelligence (AI)-based fundus screening systems have shown that they can detect five common ocular diseases, specifically diabetic retinopathy (DR), retinal vein occlusion (RVO), and pathologic retinopathy (RVO). A good effect was obtained for myopia. To new research.1
A research team from Sun Yat-sen University found that the sensitivity, specificity, precision, positive predictive value (PPV), and negative predictive value (NPV) of three fundus abnormalities were greater than those of age-related macular degeneration (>80%). suggested that (AMD), associated glaucoma, and other abnormalities.
“Epidemiologic data from ophthalmic clinics reveal that certain patients with optic nerve atrophy are young and free of ophthalmic and systemic disease. This is likely indicative of regional epidemiological characteristics, and it is difficult to identify the cause.” needs further research,” the research team wrote.1
The team led by Mr. Xuanwei Liang and Mr. Rongxin Chen of Zhongshan Eye Center chose the AI screening system in eye imaging examination for reasons such as the ease of use of the AI screening system, the limited resources required, and the appropriateness of use in key areas. suggested the possibility of clinical application. According to the literature, AI may be a future trend in fundus abnormality screening, helping to provide early diagnosis and treatment options for patients.2 Liang et al. used an AI screening system to classify 7 conditions, including 5 common fundus abnormalities, and evaluated the applicability and efficacy of the system and its application to primary population screening.
Researchers collected 637 color fundus images of 327 patients from eye clinics to evaluate clinical applications of AI. A total of 20,355 color fundus images of 10,437 of his subjects from health screening centers were then incorporated to assess screening in the general population. Only one image of her from each eye examined was used for the analysis.
When qualified color fundus images are fed into the AI software, the system can output conditions including normal, age-related macular degeneration (AMD), DR, RVO, referable glaucoma, pathologic myopia, and other abnormalities . Gold standard referrals for any condition were evaluated by her two chief ophthalmologists with over 20 years of experience. Clinical application review metrics include efficacy, precision, and application efficacy as measured by positive predictive value (PPV) and negative predictive value (NPV).
A total of 158 men (48.32%) and 169 women (51.68%) were included in the clinical setting analysis, with a mean age of 48.18 years. At population screening, the analysis included 5,481 males (52.52%) and 4,956 females (47.48%), with a mean age of 46.31 years.
According to the data, AI diagnoses included 436 normal eyes, 29 eyes with AMD, 16 eyes with DR, 17 eyes with RVO, 74 eyes with referable glaucoma, and pathologic myopia. were included in 48 eyes and other abnormalities in 17 eyes. On the other hand, according to the gold standard diagnosis, 391 eyes were normal, 14 had AMD, 17 had DR, 14 had RVO, 55 had glaucoma, 44 had pathological myopia, and 102 had other abnormalities. there were.
On the other hand, the AI population screening diagnostic results included 15,779 normal eyes, 653 AMD, 713 DR, 248 RVO, 2146 glaucoma, 350 pathological myopia, and 466 other abnormalities were included. Investigators then compared the percentages of various diagnostic conditions in the two application environments.
Regarding AI validity, the analysis showed that RVO had the highest sensitivity, lowest false negative rate and negative likelihood ratio, and DR had the highest positive likelihood ratio. In terms of precision, RVO showed the highest precision and strongest kappa values. For application effects, the researchers found that RVO had the highest NPV and other abnormalities had the highest PPV.
Liang et al. noted similar rates of various symptoms in AMD, DR, RVO, and reference glaucoma clinical application settings and population screening. On the other hand, rates of pathological myopia and other abnormalities in population screening were only slightly lower than those in eye clinics.
“All these results demonstrated the beneficial effects of our AI-based fundus system for fundus screening and epidemiological studies,” the researchers wrote.1 “There were more healthy people in physical examination centers than in eye clinics, so the proportion of normals in population screening was higher than in clinical settings.”
References
- Cao S, Zhang R, Jiang A et al. Applied effects of an artificial intelligence-based fundus screening system: clinical evaluation and population screening. Biomed Eng online. 2023;22(1):38. Published April 24, 2023. doi:10.1186/s12938-023-01097-9
- Saleh GA, Batouty NM, Haggag S, et al. The Role of Medical Imaging Modalities and AI in Early Detection, Diagnosis, and Grading of Retinal Disease: An Investigation. Biotechnology (Basel). 2022;9(8):366. Published August 4, 2022. doi:10.3390/bioengineering9080366
