Diffusion data augmentation for enhancing Norberg hip angle estimation.

Area of Science:

• Veterinary Medicine
• Medical Imaging
• Artificial Intelligence

Background:

• Canine hip dysplasia (CHD) is a prevalent hereditary orthopedic disorder in dogs.
• Accurate evaluation of hip joint conformation, using metrics like the Norberg angle (NA), is crucial for diagnosing CHD.
• Current automated NA quantification methods require manual veterinary input, limiting efficiency.

Purpose of the Study:

• To develop an automated tool for predicting the Norberg angle directly from radiographic images, eliminating the need for veterinary intervention.
• To address challenges in acquiring diverse, high-quality annotated datasets for training diagnostic models.
• To enhance the accuracy and efficiency of canine hip dysplasia diagnosis through advanced image analysis.

Main Methods:

• Utilized diffusion models to augment a dataset of 219 canine hip radiographs to 1493 images, increasing diversity and scale.
• Developed a model to predict key anatomical points (femoral head centers, acetabular edges) and radii for NA calculation.
• Evaluated the performance of 18 pretrained ImageNet models on original and augmented datasets.

Main Results:

• Incorporating generated images from diffusion models led to a significant improvement in Norberg angle estimation accuracy.
• An average improvement of 35.3% was observed based on mean absolute percentage error when using augmented data.
• The study demonstrated enhanced model performance across 18 evaluated ImageNet models post-data augmentation.

Conclusions:

• Diffusion modeling is a promising technique for expanding training datasets in veterinary medical imaging.
• Augmented data significantly enhances the accuracy of automated Norberg angle estimation for canine hip dysplasia diagnosis.
• The developed tool has the potential to streamline diagnostic workflows and improve the early detection of CHD in dogs.