Validation of Automated 3D MRI Knee Cartilage Measurements.

Rationale

AND

Objectives

To assess the agreement between automated cartilage thickness measurements derived from three-dimensional (3D) dual-echo steady-state (DESS) magnetic resonance imaging and manual radiologist measurements.

Materials And Methods

In 111 patients, cartilage thickness measurements across 21 knee subregions were compared between an automated segmentation software and manual radiologist measurements on 3D DESS and proton density (PD) fat-suppressed sequences. Agreement was assessed using Bland-Altman analysis, Pearson correlation, and intraclass correlation coefficient (ICC). A post-hoc calibration approach was applied to both comparisons. Intersequence variability between PD manual and DESS manual measurements served as a benchmark.

Results

The automated method showed a systematic overestimation relative to the PD manual (bias +0.7 mm) and the DESS manual (bias +0.9 mm). After calibration, bias was eliminated. Postcalibration ICC improved substantially (PD: from 0.29-0.61; DESS: from 0.18-0.67), reaching levels comparable to the intersequence benchmark (ICC = 0.73). Per-region analysis revealed the greatest variability in patellar and lateral tibial subregions.

Conclusion

Following a simple global calibration, automated cartilage thickness measurements achieve overall acceptable agreement with manual reference measurements that is comparable to inherent intersequence variability. This supports the clinical acceptability of the automated approach as an efficient and reproducible alternative to manual segmentation.