Computer scientists, rheumatologists and immunologists have pooled skill sets to develop an artificial neural network that can distinguish arthritis type on CT scans of the hand—rheumatoid vs. psoriatic—while also recognizing healthy joints with no arthritis at all.
The team’s report is posted online in Frontiers in Medicine: Rheumatology.
First author Lukas Folle, last author Arnd Kleyer and colleagues at Friedrich-Alexander University of Erlangen in Germany describe their work training and testing the model on 932 scans from 617 patients who were imaged with high-resolution peripheral quantitative CT (HR-pQCT).
The network proved best at using bone shape to identify healthy controls (82% accuracy), followed by rheumatoid arthritis (75%) and psoriatic arthritis (68%). Further, when fed images of joints with undifferentiated arthritis, the AI classified 86% as rheumatoid arthritis, 11% as psoriatic arthritis and 3% as healthy.
Commenting on this latter finding, the authors state that neural networks fed with less well-defined conditions such as undifferentiated arthritis could “allow assigning and clustering such conditions, which in the future and with ongoing refinement of networks could improve disease classification, i.e., in the absence of classical biomarkers.”
Folle et al. acknowledge as a limitation their use of HR-pQCT, as it’s not widely used outside of research settings.
In addition, the study excluded osteoarthritis, as the focus was on autoimmune arthritis only.
The researchers compensated for the modality’s clinical uncommonness by including a heat-map component against which the team could compare the AI. For example:
[I]n psoriatic arthritis, the corresponding hotspots [on the heat map] are located in the area of the articular entheses which have been described as articular-entheseal organs previously. Thus, we emphasize to pay attention to these apparently very specific bone alterations especially in this region using other imaging modalities, which are broadly available such as ultrasound. … Since sonography has a very high resolution, especially at the bone surface, we assume that the changes are also comparably visible. … We seek to validate our finding in a follow-up ultrasound study.”
In a news release sent by the university in May, Folle adds that the team is “very satisfied with the results of the study as they show that artificial intelligence can help us to classify arthritis more easily, which could lead to quicker and more targeted treatment for patients. However, we are aware of the fact that there are other [disease] categories that need to be fed into the network. We are also planning to transfer the AI method to other imaging methods such as ultrasound or MRI, which are more readily available.”
The journal has posted the study in full for free.
Reference:
Lukas Folle, Arnd Kleyer et al., “Deep Learning-Based Classification of Inflammatory Arthritis by Identification of Joint Shape Patterns—How Neural Networks Can Tell Us Where to “Deep Dive” Clinically.” Frontiers in Medicine: Rheumatology, March 10, 2022. DOI: https://doi.org/10.3389/fmed.2022.850552
More coverage of arthritis imaging:
Scientists develop new imaging measure to improve knee osteoarthritis staging
FDA clears artificial intelligence algorithm for diagnosing osteoarthritis on knee X-rays
New 3D imaging algorithm detects changes in arthritic joints better than x-rays
Dave P. has worked in journalism, marketing and public relations for more than 30 years, frequently concentrating on hospitals, healthcare technology and Catholic communications. He has also specialized in fundraising communications, ghostwriting for CEOs of local, national and global charities, nonprofits and foundations.