A readily available artificial intelligence system could help rheumatologists screen for capillary changes that affect the prognosis of patients with the autoimmune disease systemic sclerosis (Ssc), research suggests.
The open-source computer algorithm Vision Transformer (ViT) could detect abnormalities on nailfold capillaroscopy (NFC) images, according to findings presented at the annual meeting of the American College of Rheumatology ACR Convergence 2022.
These images are used in routine clinical practice to assess microcirculation and detect pathologic changes in thickened skin at the base of the nails.
Although tests showed that four rheumatologists generally performed better than the computer model in assessing images, except in one instance, the researchers maintain the system could aid clinicians in streamlining image assessment.
“Our study gave strong signals that the model was a reliable instrument to assess NFC images and could easily serve as an assisting device for detecting NFC changes,” said investigator Alexandru Garaiman, a rheumatology resident at University Hospital, Zurich, Switzerland.
“We see it coming into play when the rheumatologist has to assess a plethora of NFC images. This should red flag the images with abnormalities so the rheumatologist only has to assess a few images, not all of them. And at this, we think our model is doing a superb job.”
Ssc is a rare connective tissue and rheumatic disease that involves tightening and hardening of the skin and can also cause damage to blood vessels, internal organs, and the digestive system.
The researchers used NFC images from nearly 300 adults enrolled in the European Scleroderma Trials and Research Group (EUSTAR) and Very Early Diagnosis of Systemic Sclerosis (VEDOSS) local registries.
Participants had visited a physician between 2012 and 2021 and met 2013 guidelines for established systemic sclerosis, or the preliminary criteria for VEDOSS. All available images of both hands were included.
The ViT model was first trained to identify enlarged capillaries, giant capillaries, capillary loss, microhemorrhages, and abnormal morphology before being tested on new images.
Its performance was assessed with cross-fold evaluation, which is a resampling method to test and train the machine learning system, and compared with that of four rheumatologists, three of whom were trained annotators and one a treating physician.
Overall, 17,126 NFC images were assessed in 234 EUSTAR and 55 VEDOSS patients.
The ViT model performed well in identifying NFC changes in all five folds. The area under the receiving operating characteristic curve (AUC) ranged from 81.8% to 84.5% for identifying different microangiopathic changes and was 83.2% for detecting the presence of scleroderma pattern, in which there was extremely low capillary density or giant capillaries present.
In a reliability set of 464 images, the algorithm performed best at diagnosing giant capillaries and enlarged capillaries, with AUCs of 92.6% and 90.2%, respectively.
Good AUCs were seen in depicting capillary loss and microhemorrhages, at corresponding AUCs of 86.7% and 85.0%.
The rheumatologists generally performed better in assessing NFC images, although the model outperformed one in classifying the capillary loss.
“We have always seen our ViT model as a possible instrument to assist rheumatologists in screening for SSc-related microangiopathy,” said Garaiman.
“We never aimed to outperform human assessors or to replace them, but rather ease their reporting of NFC images.”
