AUTOMATIC HIGH-RESOLUTION CARTILAGE THICKNESS MAPPING AND SCORING: EXTENSION TO STANDARD CLINICAL MRI - DATA FROM THE LAUSANNE KNEE STUDY (LKS).

INTRODUCTION: A fully automated MRI-based framework was recently developed to assess femorotibial cartilage damage severity using anatomically standardized cartilage thickness maps (CTh-Maps) and an associated pattern-based severity score (CTh-Score, ranging from 0, indicating healthy cartilage, to 100, indicating end-stage osteoarthritis). The framework showed excellent reproducibility, strong agreement with expert MRI cartilage-loss assessment, and sensitivity to structural osteoarthritis progression. It was developed and validated using high-resolution DESS MRI, which is not routinely acquired in clinical practice. Standard clinical knee MRI more commonly uses tri-planar 2D fat-suppressed intermediate-weighted fast spin echo (FSE FS IW) sequences, which have lower through-plane resolution and reduced cartilage-bone contrast (Figure 1). Translating CTh-Maps and CTh-Score computation to standard clinical MRI would support routine use. OBJECTIVE: To adapt the framework to routine clinical tri-planar 2D FSE FS IW MRI and assess whether clinical MRI-derived CTh-Maps and CTh-Scores agree with DESS-based reference measurements. METHODS: This study included 430 participants of the Lausanne Knee Study (LKS), with one randomly selected knee per participant. Each underwent 3D DESS MRI, with 0.63 mm isotropic voxels, and tri-planar 2D FSE FS IW MRI, with through-plane voxel dimensions 3.3 mm. DESS cartilage segmentations generated by expert musculoskeletal radiologists served as reference standards. A 3D U-Net implemented with nnU-Net was trained to segment cartilage from interpolated and concatenated sagittal, coronal, and axial FSE FS IW images using DESS-derived segmentations as ground truth. Training included 352 subjects, with 78 held out for testing. CTh-Map accuracy, precision, and agreement were assessed using mean thickness difference, standard deviation of thickness difference, and voxel-wise ICC, respectively. CTh-Scores from FSE FS IW- and DESS-based maps were compared using ICC and Bland-Altman limits of agreement. DESS test-retest reproducibility was assessed in 11 independent knees. RESULTS: The cohort included 192 males and 238 females, with a mean age of 47.8 ± 16.2 years old . In the DESS test-retest analysis, CTh-Maps showed an accuracy of -0.01 ± 0.02 mm, precision of 0.21 ± 0.02 mm, and excellent reliability, with an ICC of 0.96 ± 0.02. In the 78-subject test set, FSE FS IW-based CTh-Maps closely agreed with DESS-based reference maps, with an accuracy of 0.04 ± 0.08 mm, precision of 0.21 ± 0.11 mm, and ICC of 0.95 ± 0.02. CTh-Scores computed from FSE FS IW- and DESS-based maps also showed excellent agreement, with an ICC of 0.91. Bland-Altman analysis showed a bias of -1.8 points on the 0-100 CTh-Score scale, with 95% limits of agreement from -7.5 to 5.0. CONCLUSION: CTh-Maps and CTh-Scores can be reliably derived from routine clinical tri-planar 2D MRI. Despite the lower through-plane resolution and reduced cartilage-bone contrast of clinical sequences, FSE FS IW-based CTh-Maps and CTh-Scores showed excellent agreement with DESS-based references (Figure 2), comparable to the DESS test-retest results. These findings support the clinical translation of automated cartilage thickness mapping and scoring to standard knee MRI protocols.