Knee osteoarthritis (KOA) is one of the worst varieties of arthritis. If not treated right away, it might lead to knee replacement. For this reason, early KOA detection is crucial for optimal therapy. This work tested and improved deep learning (DL) algorithms for predicting and identifying KOA. The suggested approach was evaluated utilizing an available dataset that included preprocessing approaches such as scaling and normalization. The Google-BER-LSTM hybrid model was explicitly designed to improve classification accuracy. The proposed binary optimization approach and other comparable methods include the Al-Biruni Earth Radius (BER), Harris Hawks Optimizer (HHO), JAYA Optimization Algorithm (JAYA), Satin Bowerbird Optimizer (SBO), Gravitational Search Algorithm (GSA), Stochastic Fractal Search (SFS), Multi-Verse Optimization (MVO), Biogeography-Based Optimizer (BBO), Whale Optimization Algorithm (WOA), Particle Swarm Optimization (PSO), and Thyroid Stimulating Hormone (TSH). The statistical study utilized ANOVA and Wilcoxon signed-rank tests to evaluate the efficacy and relevance of the suggested procedure to the ten additional methods. Furthermore, various visual representations were produced to demonstrate the suggested algorithm’s efficacy and resilience. As a result, the Google-BER-LSTM algorithm outscored the other optimizers on the bulk of the unimodal benchmark functions. For categorization, two machine learning (ML) models were utilized: multilayer perceptron (MLP) and long short-term memory (LSTM) network. The LSTM model had the best precision (PPV) of 0.9386792, negative predictive value (NPV) of 0.970845481, F-Score of 0.945368171, accuracy of 0.958558), sensitivity of 0.95215311, specificity of 0.973023881, and time of 428.4418 s. Thus, LSTM acted as a fitness function, with binary Al-Biruni Earth Radius (bBER) being used to optimize it. Finally, utilizing the suggested approach, KOA classification accuracy reached 0.995260664.
Diab et al. (Tue,) studied this question.