Diagnosis, Classification, and Assessment of the Underlying Etiology of Uveitis by Artificial Intelligence: A Systematic Review

Abstract

Recent years have seen the emergence and application of artificial intelligence (AI) in diagnostic decision support systems. There are approximately 80 etiologies that can underly uveitis, some very rare, and AI may lend itself to their detection. This synthesis of the literature selected articles that focused on the use of AI in determining the diagnosis, classification, and underlying etiology of uveitis. The AI-based systems demonstrated relatively good performance, with a classification accuracy of 93-99% and a sensitivity of at least 80% for identifying the two most probable etiologies underlying uveitis. However, there were limitations to the evidence. Firstly, most data were collected retrospectively with missing data. Secondly, ophthalmic, demographic, clinical, and ancillary tests were not reliably integrated into the algorithms' dataset. Thirdly, patient numbers were small, which is problematic when aiming to discriminate rare and complex diagnoses. In conclusion, the data indicate that AI has potential as a diagnostic decision support system, but clinical applicability is not yet established. Future studies and technologies need to incorporate more comprehensive clinical data and larger patient populations. In time, these should improve AI-based diagnostic tools and help clinicians diagnose, classify, and manage patients with uveitis.

Keywords: Bayes’ theorem; algorithms; artificial intelligence; decision support system; diagnosis; etiology; uveitis.

Figure 1

PRISMA flow diagram.

Figure 2

Risk of bias and applicability concerns table (QUADAS-2) [14]. Legend: Red (-) = high risk of bias; Yellow (?) = unknown risk of bias; Green (+) = low risk of bias [3,10,11,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35].