With artificial intelligence, early pterygium diagnosis is possible using only smartphone images, according to a report published in the British Journal of Ophthalmology. The technology could serve those who live in remote locations with less access to ophthalmology offices with dedicated equipment.
Researchers integrated slit-lamp and smartphone photographs to build effective fusion training sets for an artificial intelligence model to detect and grade pterygium.
The study’s model is built with 20,987 clear, slit-lamp eye panoramas, in addition to a second dataset of 1094 high-quality images taken with various smartphones. Photos included normal corneas, pterygium, and other ocular abnormalities. The 2 datasets were randomized; 70% to a training set, and 30% for the test set. More than 40% of the test set comprised a validation set, and training/validation pictures were not used in the test set.
The artificial intelligence early pterygium detection model trained with slit-lamp photos averaged 95.24% accuracy to identify pterygium in phone-based snapshots. For grading, the fusion segmentation model trained with images from slit-lamp and smartphones in a ratio of 83:17, respectively, proved optimal. This model reached 0.8709 sensitivity, 0.9668 specificity, and 88.31% accuracy. To support the data, researchers gathered 104 pairs of same-patient slit-lamp and smartphone photos. Using these, the fusion model analyzed cell pictures, yielding 0.9360 sensitivity, 0.9613 specificity, and accuracy at 92.38% — similar effectiveness to grading using segmentation with slit-lamp-only data that attained 94.29% accuracy.
To additionally test the model, smartphone images including 37 normal, 90 pterygium, and 73 other abnormalities were provided to 3 experienced ophthalmologists who achieved 100% detection, compared with the artificial intelligence’s early pterygium detection, which was identified 98.50% of the time. Further, the specialists graded the pterygium photographs with 93.91% accuracy, compared with the fusion artificial intelligence at 88.52%.
Prior studies have determined worldwide pterygium prevalence at 12%, ranging from 0.07% in Saudi Arabia to 53.0% in China. Also, complications are proven best managed when pterygium is caught early on, and there is less post-surgical recurrence risk when lesion tissue is small.
“Regrettably, as a specialised medical device relying on professional medical staff, the slit lamp is not always available in primary hospitals. Instead, the progress of informatisation in society endows smartphones with superb portability and universality in the general population,” the researchers report.
A limitation of this study was reliance on images only, without adding medical history and symptom information. Also, exclusion of low-quality data was done manually rather than by algorithm. However, the new model establishes a smartphone photo database, and may be useful for early diagnosis when slit-lamp screening is at a distance from rural patients.
Liu Y, Xu C, Wang S, et al. Accurate detection and grading of pterygium through smartphone by a fusion training model. Br J Ophthalmol. Published online March 1, 2023. doi:10.1136/bjo-2022-322552
This article originally appeared on Ophthalmology Advisor