Pre-trained convolutional neural network with transfer learning by artificial illustrated images classify power Doppler ultrasound images of rheumatoid arthritis joints
These videos have been matched automatically. Contact us if they are not relevant.
These videos have been matched automatically. Contact us if they are not relevant.
View abstract on PubMed
Summary
This summary is machine-generated.Artificial joint ultrasound images improved the classification performance of a convolutional neural network (CNN) for rheumatoid arthritis (RA) synovitis detection. This novel approach shows potential for medical imaging where real data is scarce.
Area Of Science
- Medical Imaging
- Artificial Intelligence in Medicine
- Rheumatology
Background
- Rheumatoid arthritis (RA) diagnosis relies on clinical assessment and imaging.
- Collecting sufficient high-quality ultrasound images for training AI models can be challenging.
- Artificial images offer a potential solution to data scarcity in medical AI.
Purpose Of The Study
- To evaluate the classification performance of a pre-trained convolutional neural network (CNN) using artificial joint ultrasonography images.
- To assess the effectiveness of transfer learning with artificial data for rheumatoid arthritis (RA) detection.
- To explore the utility of artificial images as a supplement or alternative to real clinical images in AI model training.
Main Methods
- A retrospective study utilized 870 artificial joint ultrasound images generated based on the European League Against Rheumatism/Outcome Measure in Rheumatology scoring system.
- The Visual Geometry Group (VGG)-16 CNN model underwent transfer learning, initially trained on artificial images, then refined with original and additional real images.
- Model performance was evaluated using 156 real joint ultrasound images from 74 RA patients, focusing on synovial vascularity classification.
Main Results
- The initial CNN model demonstrated moderate classification performance, with a low area under the curve (AUC) of 0.59 for grade 1 synovitis.
- The refined model, trained with additional real images, showed improved classification for grade 1 synovitis, achieving an AUC of 0.73.
- The study highlights the impact of training data augmentation on CNN performance in medical image analysis.
Conclusions
- Artificial ultrasound images can effectively augment training datasets for CNNs like VGG-16 in RA diagnosis.
- This novel method of using artificial images presents a viable strategy for medical imaging AI development, especially when real-world data collection is difficult.
- The findings suggest broader applicability in medical fields facing similar data acquisition challenges.
These videos have been matched automatically. Contact us if they are not relevant.
These videos have been matched automatically. Contact us if they are not relevant.
