Figure 5
T2 quantification performance of optimal ROI-specific pipeline in lumbar spine intervertebral discs. (a) Visual pipeline performance within the lumbar spine IVDs for a representative patient, with corresponding NRMSEs for IVDs in the predicted T2 map slice. Predicted maps are masked using an IVD segmentation mask and superimposed on the ground truth, fully sampled TE = 0 ms MAPSS echo time image. Network performance is best through R = 6, after which local T2 elevations are diffuse and underestimated. (b) Bland–Altman plots for all scans within test set (n = 5, 5 IVDs plotted for each if segmentation of disc available). T2 value predictions reflect some bias and fairly wide limits of agreement, particularly above R = 4. These results indicate progress but the need for improvement. Smaller lumbar spine dataset and test set size are likely responsible for poorer model when compared to hip and knee performance, as well as the relatively smaller number of slices in kz, which exacerbates undersampling effects.
