Hip joint image quality screening based on the Diffusion Mamba model.

Area of Science:

• Orthopedics
• Artificial Intelligence
• Medical Imaging

Background:

• High-quality annotated datasets are crucial for developing reliable AI in diagnosing hip joint pathologies.
• Existing datasets often lack the diversity and quality needed for robust AI model training.

Purpose of the Study:

• To introduce Hip-35, a large-scale, high-quality dataset of 2D hip radiographs for AI-driven orthopedic abnormality diagnosis.
• To establish a novel three-stage quality control framework for generating and refining synthetic medical images.

Main Methods:

• Developed a three-stage quality control pipeline: chromatic artifact detection, multi-metric deblurring, and anatomical hashing for deduplication.
• Utilized Diffusion Mamba (DiM) for efficient generation and refinement of synthetic hip radiograph candidates.
• Incorporated expert-annotated clinical radiographic examinations and rigorous validation processes.

Main Results:

• The Hip-35 dataset comprises 35,000 synthetic images across 35 pathology categories and 76 expert-annotated cases.
• The quality control framework achieved a 92% acceptance rate in radiologists' blinded validation.
• The dataset includes diverse clinical variations, such as fractures, deformities, and post-surgical cases.

Conclusions:

• Hip-35 provides a benchmark for medical image augmentation, facilitating reproducible AI development for rare orthopedic conditions.
• The publicly released dataset and open-source quality toolkit will advance AI diagnostic capabilities in hip pathology.
• This work addresses the scarcity of high-quality data, paving the way for more reliable AI in orthopedic diagnostics.