Deep learning AI model reliably predicts progression to early RA

Presented by

Dr Yanli Li, Leiden University Center, the Netherlands

Conference

EULAR 2023

Researchers developed a deep learning artificial intelligence (AI)-method that automatically analyses extremity MRI scans by pre- and post-processing images. After the training, the system achieved accuracies close to experts in predicting early rheumatoid arthritis (RA).

Predicting early RA from MRI images can help initiate prompt treatment, possibly preventing the chronicity of the disease. Until present, visual scoring (e.g. with the RAMRIS scoring system) from extremity MRI scans is used to manually identify key risk factors for the chance of developing RA. The study group of Dr Yanli Li (Leiden University Center, the Netherlands) assessed whether AI interpretations of MRI images could provide more accurate predictions than visual scoring by medical staff.

The model was first trained to understand anatomy from MRIs of wrists and metacarpophalangeal joints of healthy controls. In a second step, it learnt to distinguish between the different groups (patients with clinically suspect arthralgia (CSA) versus healthy controls and early-onset arthritis versus healthy controls). In a third step, the AI was taught to distinguish RA from other arthritides. Finally, the system had to predict RA development in 2 years in patients with CSA. The model’s accuracy was evaluated with the area under the receiver operator curve (AUC).

The AI analysed MRI scans from 1,974 people with either early-onset arthritis or CSA, of whom 651 went on to develop RA. On the test set, the proposed model obtained a mean AUC of 0.683 in the early-onset arthritis group and 0.727 in the CSA group. These accuracies are close to the expert level using RAMRIS.

As Dr Li emphasised, the system can be further improved with more clinical data. Moreover, the self-learning AI system showed similar efficacy for scans of either wrists or feet. As Dr Li explained during the presentation, AI-based RA prediction is reliable as it looks at known inflammatory signs and features such as synovial inflammation.

Li Y, et al. Exploring the Use of Artificial Intelligence in Predicting Rheumatoid Arthritis, Based on Extremity MR Scans in Early Arthritis and Clinically Suspect Arthralgia patients. OP0002, EULAR 2023, 31 May–3 June, Milan, Italy.

Related Articles