A statistical approach for high order epistasis interaction detection for prediction of diabetic macular edema

Abstract

Diabetic macular edema (DME) is type of common vision loss threatening disease caused due to the accumulation of fluid in the macula, a vital part of the retina that controls the most specific vision abilities. Epistasis is the phenotypic effect of one gene (which are not alleles) that is masked by a different gene (i.e.) where the influence of one gene is dependent on the presence of one or more modifier genes. Epistasis learning aims at detecting the association between multiple Single Nucleotide Polymorphisms (SNP) and complex disease. The detection of epistatic interaction helps to detect diabetic retinopathy in the used DME dataset. A method using statistical approaches like symmetric uncertainty, interaction weight factor, and interaction gain for the detection of epistasis interaction is proposed here and is compared with other prediction methods like FHSA-CED, EACO and MACOED and also with two machine learning techniques, Support vector machine(SVM) and k-nearest neighbors (kNN). The proposed method helps to predict the retinal dysfunctions earlier and prevent vision loss. Prediction of diabetic retinopathy symptoms by analysing epistatic interaction and use statistical methods for detecting the high order epistatic interaction with minimized computational complexity. Recall and the precision value of the proposed method observed is 99 and 73.880, respectively. Analysis of the accuracy and the F1 score is 99.99% and 84.615%, respectively, when compared to existing methodologies. The outstanding performance of proposed technology is observed and analyzed, and the capability of detecting the high order epistatic interaction with less computational complexity is been perceived.

Keywords

• Exudates segmentation
• Diabetic macular edema
• Single nucleotide polymorphisms (SNP)
• Diabetes
• Diabetic retinopathy
• Machine learning