A benchmarking platform for selecting optimal retinal diseases diagnosis model based on a multi-criteria decision-making approach
Neven Saleh, Ahmed M. Salaheldin
Abstract
Many automated diagnosis models are designed to classify retinal disease using the Optical Coherence Tomography images. The challenge is how to benchmark such diagnosis models to enable the decision-maker to select the superior diagnosis model. This problem is rarely tackled in the literature. This study aims to evaluate and benchmark the automated diagnosis models of retinal disease to select a superior model. Multi-Criteria Decision-Making-based platform has been proposed to address this issue employing a set of nine quantitative criteria. By examining the underlined research, and according to the proposed criteria, only five models have been selected for the benchmarking. The platform has been designed using the Entropy-Technique for Order Preference by Similarity to Ideal Solution (Entropy-TOPSIS) method. The Entropy is used to find the weights of the criteria; meanwhile, the TOPSIS is implemented to find the ranking order of the models. The results of benchmarking yielded that the EfficientNet-B7 Convolutional Neural Network is the superior classifier with a closeness coefficient of 0.999 of the data set. Adopting an efficient platform for classifying retinal disorders promotes a reliable and fast diagnosis model in terms of reducing human errors and saving time. Furthermore, the study highlights the significance of some new criteria.