Summary:
UCLA researchers in the Department of Electrical and Computer Engineering have developed an AI-based interface designed to enable individuals without specialized training to identify arthritis in medical imaging.
Background:
The use of artificial intelligence (AI) for medical imaging analysis holds great promise for the future of healthcare. One significant application of AI is in clinical decision-making, where it boosts the accuracy and speed of medical imaging analysis. An example of a promising new application is the imaging-based diagnosis of arthritis and degenerative joint disease (DJD), using techniques like X-ray, MRI, CT, and ultrasound. However, AI requires extensive training data consisting of medical images annotated with their diagnoses. Obtaining such data typically involves hiring domain experts, such as radiologists, which can be prohibitively expensive and limit access to high-quality, labeled datasets. Traditional approaches are constrained by economic costs associated with hiring specialists, thereby restricting the availability of qualified data for AI applications in diagnostic imaging. This bottleneck creates a critical need for cost-effective, scalable methods to generate high-quality labeled datasets without compromising diagnostic accuracy. Addressing this gap could accelerate the development and deployment of AI in specialized diagnostic areas such as arthritis imaging.
Innovation:
Researchers at the UCLA Department of Electrical and Computer Engineering have developed an AI-based interface that enables individuals without specialized expertise to diagnose medical imaging for arthritis at a level comparable to experts. The system promotes critical analytical thinking by guiding users through a structured process that clarifies the relationships between key diagnostic criteria and the specific features of arthritis. The interface consists of two primary phases: Criteria Phase and Correction Phase. In the Criteria Phase, users review an image, identify key diagnostic criteria, annotate areas of interest, and record an initial diagnosis based on existing medical guidelines. The Correction Phase refines these annotations through four steps: the system proposes a diagnosis and highlights areas of interest, users compare their annotated areas to AI-generated areas, a quantitative overlap analysis displays the degree of correspondence between the two sets of criteria, and users finalize the diagnosis as a training label. In a controlled user study involving 12 non-experts, the system demonstrated that users could achieve labeling accuracy comparable to experts, particularly for images with low uncertainty scores (a marker of sub-optimal AI performance). This innovation, titled DANNY (Data ANnotation for Non-Experts made easY), addresses a critical need for accurate, cost-effective dataset labeling by non-experts, promoting broader AI implementation in data-intensive domains.
Potential Applications:
• Diagnostic support for arthritis in primary care settings
• Educational tool for medical students to practice diagnostic criteria identification
• Integration into telemedicine platforms
• Dataset labeling for training AI models in arthritis detection and classification
• Quality control in medical imaging analysis to reduce diagnostic variability
Advantages:
• Increasing diagnostic accessibility for non-experts in underserved areas
• Reduces reliance on specialized experts for diagnosis, lowering costs
• Enhances diagnostic accuracy through iterative feedback and correction phases
• Facilitates dataset creation with expert-level labeling accuracy for AI training
• Promotes analytical thinking by guiding users through structured diagnostic criteria assessment.
State of Development:
A paper describing this invention has been submitted for publication and was presented on March 1, 2025.
Reference:
UCLA Case No. 2025-108
Lead Inventor:
Xiang Anthony Chen, UCLA Associate Professor of Electrical and Computer Engineering.