Fully Automatic Knee Joint Segmentation and Quantitative Analysis for Osteoarthritis from Magnetic Resonance (MR)...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Knee osteoarthritis (OA) diagnosis relies on accurate imaging biomarkers.
• Manual segmentation of knee MRIs is time-consuming and subjective.
• Developing automated methods is crucial for efficient OA assessment.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) method for automated knee joint MR image segmentation.
• To quantitatively compute knee OA-related imaging biomarkers using DL.
• To assess the correlation of automatically computed biomarkers with manual segmentation and OA progression.

Main Methods:

• A retrospective study utilized 843 knee MR imaging volumes.
• A convolutional neural network (CNN) with multiclass gradient harmonized Dice loss was employed for segmentation.
• Morphologic biomarkers including cartilage/meniscus volume and thickness, and minimal joint space width (mJSW) were computed and compared.

Main Results:

• The CNN model achieved high Dice coefficients for bone compartments (0.948-0.974), cartilage (0.717-0.809), and menisci (0.846).
• Automated biomarkers strongly correlated with manual segmentation (ICCs: 0.916-0.876).
• Cartilage measurements and mJSW showed strong correlation with knee OA progression.

Conclusions:

• A fully automatic CNN-based knee segmentation system was developed for efficient knee joint image evaluation.
• The system reliably computes and visualizes OA-related biomarkers.
• This AI tool can assist radiologists and orthopedic surgeons in clinical practice and research for knee OA assessment.