Detection of knee osteoarthritis degree via optimized deep learning architectures

Osteoarthritis (OA), a chronic degenerative joint disorder, can occur in different parts of the body for various reasons. As the most common type of disease, knee OA (KOA) causes both a negative impact on public health and a serious burden on the social economy of which grading constitutes a challenging issue on X-ray images. In this paper, to handle this issue, the optimized transfer learning-based models are extensively examined which constitute the main aim of our paper on both multi- and binary-categorization tasks. The hyperparameter arrangement of transfer learning (TL) models is handled as a discrete & continuous optimization problem, and state-of-the-art optimization methods are chosen and compared to efficiently solve this competitive – NP-hard problem. Sixteen optimized architectures are designed using four influential optimization methods (ASPSO, CDW-PSO, CSA, MSGO) and four effective TL models (MobileNetV2, ResNet18, ResNet50, ShuffleNet) to classify the X-ray KOA images. Regarding the experiments on both binary and multiclass categorizations, it's seen that the MSGO algorithm arises as the robust method to be considered for hyperparameter tuning of TL-based models by achieving high performance. In addition, it's seen that MobileNetV2 and ResNet-based models come to the forefront of X-ray imaging-based classification by achieving high accuracy rates due to the usage of residual blocks. Consequently, in terms of mean accuracy, ResNet50-MSGO and MobileNetV2-CSA respectively record 93.15% and 93.29% success rates on multiclass categorization, whilst ResNet18-CDW-PSO and MobileNetV2-MSGO provide the same highest score (99.43%) on binary categorization.