Monday August 28th 2023



A fully automated machine learning algorithm can now predict bone marrow inflammation, known as bone marrow edema, in MRI scans. This innovative algorithm, developed by the teams of Prof. Yvan Saeys and Prof. Dirk Elewaut at the VIB-UGent Center for Inflammation Research and colleagues at University Hospital Ghent (UZGent), offers an objective and standardized way to evaluate inflammatory lesions, and it has the potential to impact the diagnosis and treatment of spondyloarthritis patients significantly. The work appears in Arthritis & Rheumatology.

Detecting inflammation 

Traditionally, the detection of active sacroiliitis (inflammation of the sacroiliac joint that connects our hips to our spine) plays a crucial role in the early diagnosis and monitoring of inflammatory conditions such as spondyloarthritis. However, evaluating MRI scans of this joint is challenging due to the variability of inflammatory lesions and the expertise required for accurate interpretation. 

While most of the news is concerned about chatbots, AI is revolutionizing medical imaging. In this new study, researchers from VIB-UGent and UZGent developed a computer vision workflow to automate the entire process from detecting the SI joint and segmenting the regions to predicting bone marrow edema.

Dr. Joris Roels (VIB-UGent): "Our algorithm was trained and tested using a diverse dataset of arthritis patients, postpartum individuals, and healthy subjects. The predictions were further validated using an independent dataset of MRI scans from arthritis patients." 

"With that training," Dr. Ann-Sophie De Craemer (UZGent) adds, "we got promising results from our machine learning model, with roughly between 70 and 80% accuracy in detecting inflammation in the sacroiliac joint." ​ 

Assisting diagnosis 

Getting automated predictions of inflammation from three-dimensional images of a complex anatomical structure is very challenging, but by ‘segmenting’ the image into more detailed sections, this new algorithm can make the search space smaller, which leads to better predictions. ​ 

Prof. Dirk Elewaut (VIB-UGent): "We developed a fully automated machine learning pipeline that allows a standardized evaluation of sacroiliac inflammation on MRI scans. This method can potentially screen large numbers of suspected arthritis patients and brings us one step closer to using artificial intelligence for diagnosis and follow-up." 

Prof. Yvan Saeys (VIB-UGent) adds: "With further refinement and validation, this technology could revolutionize healthcare practices and improve patient outcomes. The role of AI here is not to replace, but to assist human specialists, so that they gain time for interacting with their patients. ​ Using the explainable AI features of our model, medical experts can now also get a grasp of where the AI model is looking at in the image, rendering our AI models more trustworthy and explainable."