DR. ADITYA ANAND

DR. FAUZIA ARA, DR. SOWMYA SHREE B V, DR. SHARADA M

Abstract

Purpose: To analyze the latest application of deep learning algorithm in glaucoma screening & progression using fundus photography, OCT, SAP and with structure & function.

Method: Representation learning in convolutional neural network(CNN) type deep learning model was compared with cloud data which was based on traditional machine learning using support vector machine. Back propagation & transfer learning techniques are used for analysis in all parameters.

Result: Considering individual parameters, fundus photography delivered 84.5% sensitivity, 98% specificity & ROC of 0.942. OCT parameter results have shown 90% sensitivity, 92% specificity & ROC of 0.928. An accuracy of 87% is given with SAP. Glaucoma progression analysis showed 75% sensitivity & 88% specificity. Precision of 85% with structure and function.

Conclusion: Deep learning algorithm is a recent tool for glaucoma diagnosis, progression and its management with the potential for low cost screening method at community level.

Full Text