Automatic estimation of knee effusion from limited MRI data.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Knee effusion is a common issue in osteoarthritis, often graded using semi-quantitative scales.
• Accurate quantification of knee effusion is crucial for osteoarthritis assessment and management.

Purpose of the Study:

• To develop and evaluate a dense neural network (dNN) for detecting knee effusion in MRI scans.
• To assess the dNN's performance on low-resolution images for potential use in low-cost MRI systems.

Main Methods:

• A dNN was trained using sagittal turbo-spin-echo (TSE) MR images from the Osteoarthritis Initiative (OAI) dataset.
• The dNN's accuracy was compared to VGG16 and a musculoskeletal radiologist.
• Robustness was tested by adding Gaussian noise to images.

Main Results:

• The dNN achieved an average accuracy of 62% on a large test dataset.
• The network demonstrated robustness to image noise, maintaining high accuracy.
• On a smaller dataset, the dNN outperformed a human radiologist in classifying knee effusion.

Conclusions:

• Neural networks can effectively classify knee effusion from low-resolution MRI scans.
• AI-driven assessment holds potential as a useful tool for low-cost, low-field MRI systems.
• Automated knee effusion assessment using NNs is feasible and comparable to expert human evaluation.