Self- and Cross-attention based Transformer for left ventricle segmentation in 4D flow MRI
Xiaowu Sun, Li-Hsin Cheng, Rob J. van der Geest
The conventional quantitative analysis of 4D flow MRI relies on the co-registered cine MRI. In this work, we proposed a self- and cross-attention based Transformer to segment the left ventricle directly from the 4D flow MRI and evaluated our method on a large dataset using various metrics. The results demonstrate that self- and cross-attention improve the segmentation performance, achieving a mean Dice of 82.41$\%$, ASD of 4.51 mm, left ventricle ejection fraction (LVEF) error of 7.96$\%$ and kinetic energy (KE) error of 1.34 $\mu$J$/$ml.
Thursday 7th July
Poster Session 2.2 - onsite 11:00 - 12:00, virtual 15:20 - 16:20 (UTC+2)