UDC 616.379-008.64]:617.735-005-079-052:575

Abstract. Background. It is known that in diabetic retinopathy (DR), impaired transforming growth factor β1 (TGF-β1) signaling is accompanied by pathological angiogenesis, disruption of the blood-eye barrier, activation of inflammation and tissue fibrosis. The purpose of the study was to establish the relationship between the content of TGF-β1 in blood serum and intraocular fluid (IOF) and the progression of DR in type 2 diabetes mellitus (T2DM) using neural network modeling. Materials and methods. The study included the results of the examination of 102 people with T2DM, who were divided into 3 groups according to the stages of DR: the first one — non-proliferative DR (NPDR, 35 people), the second one — preproliferative (PPDR, 34 people) and the third one — proliferative (PDR, 33 people). The control group consisted of 61 individuals. The patients underwent standard ophthalmic examinations. TGF-β1 in blood serum and IOF was evaluated by enzyme-linked immunosorbent assay (Invitrogen Thermo Fisher Scientific, USA). Statistical analysis of the results was performed using the MedCalc software package (MedCalc SoftWare bvba, 1993–2013) and a two-layer neural network model with a linear postsynaptic potential function. Results. Using the genetic selection algorithm, 3 features were identified that were associated with DR: diabetes compensation and TGF-β1 content in blood and IOF. T2DM was compensated in 38 (37.3 %) patients, while in 64 cases (62.7 %), it was uncompensated. The proportion of the latter was higher in PDR than in NPDR and PPDR (p < 0.05). In PDR, the TGF-β1 content in IOF was significantly higher than in NPDR and PPDR (p < 0.05). A three-factor classification model was created on the identified features, which included a system of equations that predicted PDR with 100% accuracy. The overall prediction accuracy of the model was 88.2 % (95% CI 80.4–93.8 %). Conclusions. In this study, the value of indicators such as diabetes compensation and TGF-β1 content in serum and IOF for the progression of DR to PDR was shown using the method of neural network modeling. Keywords: proliferative diabetic retinopathy; diabetes mellitus; transforming growth factor β1; intraocular fluid; neural network modeling