DC-AAE: Dual channel adversarial autoencoder with multitask learning for KL-grade classification in knee radiographs.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Knee osteoarthritis (OA) is a leading cause of disability in older adults, with manual assessment being subjective and variable.
• Current deep learning methods for OA grading are limited by insufficient labelled data, impacting performance and class handling.

Purpose of the Study:

• To develop a novel, fully automatic Kellgren and Lawrence (KL) grade classification system for knee radiographs.
• To leverage semi-supervised multi-task learning to enhance KL-grade classification using both labelled and unlabelled data.

Main Methods:

• A dual-channel adversarial autoencoder was trained unsupervisedly for reconstruction.
• A multi-task learning framework incorporated an auxiliary leg side identification task to utilize additional datasets.
• Semi-supervised learning exploited unlabelled data to improve feature learning for KL-grade classification.

Main Results:

• The proposed model achieved state-of-the-art performance with accuracy (75.53%), precision (74.1%), recall (78.51%), and F1 score (75.34%).
• Utilizing unlabelled data and an auxiliary task significantly improved KL-grade classification performance.
• The model demonstrated remarkable robustness across evaluations on two large public datasets.

Conclusions:

• The developed semi-supervised multi-task learning approach effectively addresses the challenge of limited labelled data in knee OA grading.
• This method enhances feature learning, leading to superior KL-grade classification performance and robustness.
• The findings suggest a promising direction for automated, objective OA assessment in clinical practice.