Swarm-Optimized Multi-Scale Deep Learning for Accurate Knee Osteoarthritis Assessment

Abstract

Knee Osteoarthritis is a weakening joint disorder, which mostly destroys the knee articular cartilage and in its cruel phases causes severe pain and frequently mains to complete joint replacement. It mainly affects aging, overweight, and those who survive in a sedentary lifestyle. Prompt analysis is vital for pathology and medical treatments. Professional radiologists visually inspect the joint of knee areas and discover the structures utilizing image processing and computer vision (CV)-based techniques, and might even utilize CAD-based extents for evaluating the severity of knee OA. However, manual approaches are highly experience-based and subjective, and they are difficult and tedious when a huge amount of issues want to be inspected. So, the machine learning and deep learning- based approaches are getting attention and distributed higher accuracy and adequate precision for the recognition and identification of KOA at an initial phase. In this study, we offer an Enhanced Multi-Scale Feature Fusion for Knee Osteoarthritis Severity Detection Using the Dwarf Mongoose Optimization (EMSFF-KOSDDMO) model. The main purpose of EMSFF-KOSDDMO model is to enhance the recognition of knee osteoarthritis severity using medical image data. Initially, the EMSFF-KOSDDMO technique applies non-local means (NLM) filtering for noise removal and adaptive gamma correction with weighting distribution (AGCWD) for contrast enhancement to enhance the knee X-ray image quality. Besides, the presented EMSFF-KOSDDMO model designs a multi-scale feature fusion of EfficientNetB0 with MBConv4 and EfficientNetV2-B0 with MBConv6 for the extraction of feature process to capture rich, multi-scale features with efficient computational performance. For the detection process, the TabNet method has been exploited. At last, the DMOA alters the hyperparameter values of TabNet system optimally and outcomes in greater performance of classification. A wide sort of experimentations are conducted to validate the performance of EMSFF-KOSDDMO system under various evaluation measures. The simulation results suggested that the EMSFF-KOSDDMO system emphasized improvement over other existing models.