Performance of Fully Automated Algorithm Detecting Bone Marrow Edema in Sacroiliac Joints

Area of Science:

• Radiology
• Artificial Intelligence
• Rheumatology

Background:

• Axial spondyloarthritis (axSpA) diagnosis relies on detecting active inflammation, often bone marrow edema (BME).
• Accurate MRI assessment of sacroiliac joints (SIJs) is crucial for axSpA evaluation.
• Technical variations in MRI acquisition can impact BME detection.

Purpose of the Study:

• To evaluate a fully automated algorithm for detecting BME in iliac and sacral bones in axSpA patients.
• To assess the algorithm's performance relative to MRI quality, specifically the coronal oblique plane deviation angle.
• To compare automated BME detection with manual segmentation.

Main Methods:

• 173 patients with suspected axSpA underwent MRI of the SIJs.
• MRI quality was assessed using a deviation angle measured in the sagittal plane.
• A fully automated algorithm performed bone and BME lesion segmentation on T1 and STIR sequences.
• Automated segmentations were compared to manual segmentations and assessed using the Dice coefficient and SPARCC scale.

Main Results:

• The algorithm achieved a high Dice coefficient (0.9820) for bone segmentation, indicating excellent agreement with manual segmentation.
• Sensitivity for BME detection ranged from 0.58 to 0.83 across different MRI quality groups, with overall sensitivity of 0.76.
• Specificity for BME detection remained high at 0.97 across all groups, demonstrating robustness to image quality variations.

Conclusions:

• The fully automated algorithm demonstrates satisfactory performance in detecting BME in axSpA patients.
• Algorithm performance is consistent and reliable, irrespective of the coronal oblique plane deviation angle (MRI quality).
• This automated approach offers a promising tool for objective and efficient BME assessment in axSpA.