A Diabetic Retinopathy Classification and Analysis Towards the Development of a Computer-Aided Medical Decision Support System

Aknan M.; Chandra S.; Gupta S.; Cotrina Aliaga J.C.; Mohapatra S. · SN Computer Science (2023) · DOI: 10.1007/s42979-023-02178-1

Resumen

Recent advances in digital image analysis have improved diagnostic procedures and given healthcare experts vital information. Previously, ophthalmologists analysed medical images, which were limited by their non-systematic search pattern, image noise, and illness complexity. CAD systems were developed by combining image data and clinical information. This technology detects lesions, assesses disease extent, and supports diagnostic decisions to improve healthcare systems. diabetic retinopathy has increased mortality. Using a computer-aided decision support system can help ophthalmologists detect diabetic retinopathy early and reduce blindness. Retinal image screening classifies and grades retinal images effectively. Decision assistance systems were developed using many methods. Using a variety of samples and experts, a computer-aided diagnostic system was developed and evaluated to categorise retinal images using SVM. It can be concluded from the results that the accuracy with which retinal pictures are categorised as normal or abnormal increases by using several criteria and selecting the SVM classifier. There is no correlation between the size of the input space and the computational complexity of SVM. Therefore, it boosts classification precision across the board by decreasing erroneous acceptance and false rejection rates. The SVM classifier has been found to have both a lower false acceptance rate and a lower false rejection rate than competing classifiers. After putting the suggested system through its paces, its accuracy was determined to be 97.65 percent. The proposed technology can act as a third party observer in clinical decision-making. Images of lesions had an average sensitivity of 96.89 percent, a specificity of 98.76 percent, and an accuracy of 98.1 percent. The experimental findings prove that the proposed technique has better sensitivity, specificity, accuracy, and predictive values than the alternatives. The proposed method produces grading outcomes that are comparable to those obtained using other methods.