Left atrium 4D-flow segmentation with high-resolution contrast-enhanced magnetic resonance angiography

Figures

• 

  Figure 1. Decision tree for subject inclusion and exclusion. Patients with 4D-flow MRI scan from CIROC database were included. After excluding apparent quality issue data, volume segmentation of the two methods was assessed. Cases were evaluated after registration. Apparently failed cases were excluded from the final analysis.

• 

  Figure 2. PC-MRA image example of excluded cases due to 4D-flow image quality issue. (A) Stripe. (B) Partly void due to low intensity. (C) Low contrast noisy image.

• 

  Figure 3. A sample case of the CE-MRA registration method. Segmentation was done on the high-resolution CE-MRA image (A,B). PC-MRA (C) was overlayed on CE-MRA, and registration was processed (E, top). CE-MRA segmentation was transformed following the same transform (E, bottom). The first round of registration may match the aortic arch rather than the heart chambers due to the difference in distance between them in the two images (E, middle column). For these cases, ROI was set around the heart chamber (D), and registration was processed once more (E, right column).

• 

  Figure 4. (A) LA segmentation from CE-MRA is strongly correlated with the segmentation from PC-MRA (r = 0.846, p < 0.001). (B) Bland-Altman plot of LA volume from two segmentation methods with the 95% limits of agreement.

• 

  Figure 5. Samples of registration. Three successful registration cases with different Dice similarity scores and one rejected case are shown here. The left column shows LA segmentations with adjacent flows. The middle column shows two segmentations only. The right column shows the pathlines of LA flow. yellow: PC-MRA LA segmentation; green: CE-MRA LA segmentation.

• 

  Figure 6. Registration result evaluation. (A) Mean of the Dice similarity coefficient between LA segmentation from PC-MRA and CE-MRA is 0.770. (B) Mean of the Average Symmetric Surface Distance was 2.704 mm.

• 

  Figure 7. Comparison of functional evaluation using LA volume segmented from PC-MRA and CE-MRA. (A) Mean velocity from two volumes was strongly correlated. (B) Bland-Altman plot of mean velocity presents no proportional bias. (C) Mean stasis from two volumes was strongly correlated. (D) Mean difference was 3.28%, and there was no proportional bias.