BACKGROUND: To evaluate Deep Learning (DL) Artificial Intelligence (AI) application to optical coherence tomography scans of non-infectious pan-uveitis patients.
METHODS: Consecutive patients with non-infectious pan-uveitis were included in the study if they had an OCT scan (Zeiss, Germany) of the macula. Inflammation was graded on the Nussenblatt scoring system by a uveitis specialist. A DL AI system developed using Python 3.9 and open source TensorFlow software was then trained to classify inflammation as high, low, or none. Performance was assessed by AUC scores. Training to cross-validation used an 80:20 split with 2000 iterations in all evaluations.
RESULTS: 154 scans of 52 patients were analyzed. 38 (70.4%) of patients were female and the mean age of participants was 50.3 ± 15.5 years. 58 (37.7%) of OCT scans were graded as no inflammation, 61 (39.6%) scans as mild inflammation, and 35 (22.7%) scans as high inflammation. The AUC for OCT images was 0.830 (CI95: 0.784 to 0.876).
CONCLUSION: Deep Learning AI can be applied in the grading of non-infectious pan-uveitis OCT scans in a precise manner.