EFRNet: A Deep Hybrid Model for End-to-End Knee Osteoarthritis Detection and Progression Assessment

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Knee osteoarthritis (KOA) diagnosis relies on X-rays, which are subjective and prone to errors.
• Current methods require expert interpretation and can be affected by image noise and low resolution.

Purpose of the Study:

• To develop and evaluate an enhanced Faster Regional Convolutional Neural Network (EFRCNN) for improved KOA detection and classification.
• To address limitations of traditional KOA assessment using digital radiography.

Main Methods:

• Implemented a hybrid pretrained module combining ResNet101, VGG19, and MobileNetV2 for superior feature extraction.
• Utilized a modified Anscombe transform to mitigate noise and low-resolution issues in X-ray images.
• Trained and validated the EFRCNN model on publicly available KOA data from the Osteoarthritis Initiative.

Main Results:

• The proposed EFRCNN model achieved a high accuracy of 98.70% in KOA detection and classification.
• The hybrid feature extraction module and modified Anscombe transform significantly enhanced diagnostic performance.
• The model demonstrated superior efficacy compared to existing methodologies.

Conclusions:

• The EFRCNN model offers a robust and accurate AI-driven solution for knee osteoarthritis diagnosis.
• This approach has the potential to enhance the reliability and efficiency of KOA assessment in clinical practice.
• Further research can explore broader applications of this enhanced deep learning technique in musculoskeletal imaging.