Purchase this article with an account.
Naoya Nezu, Yoshihiko Usui, Masaki Asakage, Hiroyuki Shimizu, Marina Ogawa, Naoyuki Yamakawa, Chihiro Yanagida, Kinya Tsubota, Akitomo Narimatsu, Katsuhiko Maruyama, Akira Saito, Masahiko Kuroda, Hiroshi Goto; Determining immune-related factors of intraocular diseases by artificial intelligence methods. Invest. Ophthalmol. Vis. Sci. 2019;60(9):5369.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Research on immune-related factors has great clinical significance in the field of ophthalmology, because analysis of immune mediators contributes to diagnosis and treatment. Previously, we simultaneously measured a broad spectrum of immune mediators including cytokines, chemokines, and growth factors in patients with several intraocular diseases. In this study, we investigated 28 immune-related factors in ocular fluids (aqueous humor and vitreous) in patients with intraocular diseases using support vector machine (SVM) compared to conventional stepwise variable selection, and canonical discriminant analysis.
670 eyes with 23 intraocular diseases [acute retinal necrosis (ARN), cytomegalovirus retinitis, vitreoretinal lymphoma, Vogt-Koyanagi-Harada disease, Behcet disease, sarcoidosis, toxoplasmosis, tuberculous choroiditis, uveitis of unknown etiology, endophthalmitis, age-related macular degeneration, retinitis pigmentosa, retinal vein occlusion, rhegmatogenous retinal detachment, proliferative diabetic retinopathy, epiretinal membrane, macular hole, uveal melanoma, metastatic tumors to the uvea, retinal hemangioma, atopic dermatitis, glaucoma, and cataract] were studied. Twenty-eight immune mediators (IL-1α, IL-1β, IL-2, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IL-10, IL-12p70, IFN-γ, TNF-α, IP-10, MCP-1, MIP-1α, MIP-1β, RANTES, Mig, VEGF, G-CSF, GM-CSF, bFGF, Fas ligand, granzyme A, granzyme B, angiogenin) were measured by Cytometric Bead Array. Data were analyzed by SVM, stepwise variable selection, and canonical discriminant analysis, and compared each other.
All three methods revealed disease-specific profiles. The overall accuracy of SVM, stepwise variable selection, and canonical discriminant analysis was 76.3%, 71.8%, and 65.7% for aqueous humor, and was 93.6%, 78.5%, and 76.5% for vitreous samples. ARN and vitreoretinal lymphoma were the diseases with high prediction accuracy using all three methods. IFN-γ, IL-10, and MCP-1 were the main cytokines for discriminating among various eye diseases such as ARN and vitreoretinal lymphoma.
SVM method utilizing intraocular fluids may be effective to predict the diagnosis of ARN and vitreoretinal lymphoma.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.
This PDF is available to Subscribers Only