Deep learning model applied for the automatic detection of joint inflammation

The Medical Delta program 'From man to machine – early identification of rheumatoid arthritis' applies a deep learning model (ADMIRA) that can fully automatically score the different aspects of arthritis (joint inflammation) from MRI images. These are the first steps in a process aimed at enabling earlier detection of rheumatoid arthritis.

This AI model, which has been trained on an extensive database of MRI scans, has now been extensively validated in patients with joint pain and joint inflammation in secondary care.

Because imaging is now performed using a new, shorter MRI sequence, initial steps have been taken to enable the model to operate on new MRI data. In this new MRI protocol, contrast agents are no longer required and scan times are significantly reduced.

In addition, software has been developed that allows the ADMIRA model to be applied immediately after an MRI scan is acquired. In a test environment with a local PACS (the regular hospital imaging system), the analysis can be initiated automatically and the results can be sent back to the PACS.

First steps toward early detection of rheumatoid arthritis

These developments represent the first steps toward creating, in the longer term, a system in which a general practitioner can request an MRI scan when joint inflammation is suspected. The AI system would then immediately analyse the images and return a report to the general practitioner.

Early recognition and intervention in Rheumatoid Arthritis (RA) can significantly reduce disease burden and associated healthcare costs.

At present, early detection is only possible in secondary care through physical joint examination by rheumatologists. This approach is inefficient, as it requires costly specialist manpower and leads to delays due to waiting lists.

Labour-saving analyses

A current local solution within the Medical Delta region, the 'Early Arthritis Recognition Clinic', has proven effective. Patients for whom general practitioners suspect arthritis are screened during a five-minute consultation involving joint examination by a rheumatologist (the so-called '1.5-line care setting'). However, this approach still requires specialist manpower and is not future-proof.

The Medical Delta program 'From man to machine – early identification of rheumatoid arthritis' aims to replace manpower with technology.

This project builds on previous research involving more than 1,700 individuals, which demonstrated that MRI is both sensitive and specific for detecting joint inflammation in secondary care. Moreover, recent advances in MR physics and image processing enabled researchers to develop a short MRI sequence as a screening tool.

Testing in daily clinical practice

The most important step in reducing manpower requirements is automated image interpretation. The Medical Delta program 'From man to machine – early identification of rheumatoid arthritis' aims to address key elements of this challenge by retraining a neural network.

“This network was developed using MRI data from secondary care. We now want to retrain it using data from the 1.5-line care setting,” says Prof. Dr Annette van der Helm (LUMC, Erasmus MC), Medical Delta professor and one of the program leaders. “In addition, we aim to integrate the AI algorithm into software used in both secondary and primary care.”

The idea is that this system can be operated by technicians with vocational or applied sciences training. After completion of this project, the pathway to implementation will be established, stimulating further research to test and optimize accuracy in daily clinical practice.

The participating parties in this Medical Delta research project are LUMC, ReumaNederland, Erasmus MC, TU Delft, and Huisartsen Campus The Hague.

Related: Medical Delta Programma ‘From man to machine – early identification of rheumatoid arthritis’| Medical Delta