Search

We applied the machine learning procedure to the analysis of serum and aqueous humor albumin, IgG and interferon-γ patterns. The data were analyzed with respect to the cataract, type of diabetes, retinopathy and blood-aqueous humor barrier damage. Intraocular production of IgG was detected in diabetic patients, since the IgG index values were increased in some diabetic groups (especially those with type I diabetes and retinopathy). Comparison of numerical methods and clonal IgG detection suggested that intraocular IgG patterns in diabetes are predominately polyclonal. Interferon-γ (IFN-γ) was pathological in serum and aqueous humor of patients with type I diabetes and to a lesser extent in the type II diabetes group. Machine learning system based on C5.0 classifier extracted 98.5% accurate rules for discrimination of the senile cataract group, diabetes group and diabetic retinopathy group. Our results confirm that the combination of immunochemical and numerical methods with a proper statistical analysis may contribute to the diagnosis of the pathologic ocular immune response and the related retinal or vascular disorders in diabetes. Significant savings in laboratory material and efforts may be obtained by means of the machine learning based optimization of the tests.