OpenMSK: Open-Source AI Framework Extracts Knee MRI and PET Biomarkers

Automated knee MRI segmentation in an open-source pipeline achieved Dice scores of 0.84–0.98 across all tissues. Quantitative PET-MRI biomarkers showed generally high agreement, but confidence intervals were wide for several. Tested on 20 prospective PET-MRI knees.

• Pipeline trained on 347 knee MRIs across 7 scanners; prospectively validated on 20 PET-MRI knees against two manual annotators, yielding Dice 0.84–0.98 with high spatial agreement.
• Quantitative biomarker agreement (cartilage T2, bone shape, PET measures) was high overall but confidence intervals were wide for some, likely reflecting the small test set.
• OpenMSK is a freely available, modular platform that lowers barriers to whole-joint quantitative imaging research.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.