Ultrafast deep learning super-resolution single-shot T2-weighted imaging for robust edema visualization in cardiovascular magnetic resonance

Figures

• 

  Fig. 1. Cardiovascular magnetic resonance imaging examples of a 62-year-old participant with follow-up of pericarditis showing standard breath-hold multi-shot T2-weighted STIR sequence (standard T2-STIR), conventional SSH T2-STIR, and DL-SSH T2-STIR acquired with breath-hold and free-breathing. STIR short tau inversion recovery, SSH T2-STIR single-shot T2-STIR, DL-SSH T2-STIR deep learning reconstructed single-shot T2-STIR

• 

  Fig. 2. Scatter plot showing mean and standard deviation, depicting significant accelerated acquisition duration (without breath-hold time) and total scan time (including breath-hold time) for DL-SSH T2-STIR with BH and FB compared to the standard T2-STIR black-blood (standard T2-STIR) sequence (all p<.0001). DL-SSH T2-STIR deep learning reconstructed single-shot T2-STIR black-blood imaging, BH breath-hold, FB free-breathing

• 

  Fig. 3. Stacked bar charts show the distribution of Likert scale scores used for qualitative image assessment of DL-SSH T2-STIR with BH and FB compared to the standard T2-STIR black-blood (standard T2-STIR) sequence. Edge definition Endocardial edge definition, DL-SSH T2-STIR deep learning reconstructed single-shot T2-STIR black-blood imaging, BH breath-hold, FB free-breathing

• 

  Fig. 4. Cardiovascular magnetic resonance imaging examples depicting artifacts in DL-SSH T2-STIR acquired with BH and FB and standard T2-STIR. (A) 66-year-old participant with sufficient breath-holding capacity; vertically oriented linear artifacts in FB DL-SSH T2-STIR, which most likely represent ghost artifacts of the chest wall due to excessive chest wall movement (yellow arrow). (B) 35-year-old participant with inconsistent breath-holding and distinctly better diagnostic quality of FB DL-SSH T2-STIR compared to both breath-holding sequences (non-diagnostic quality of standard T2-STIR). DL-SSH T2-STIR deep learning reconstructed single-shot T2-STIR black-blood imaging, BH breath-hold, FB free-breathing

• 

  Fig. 5. Stacked bar charts show the distribution of Likert scale scores used for overall qualitative image assessment of DL-SSH T2-STIR with BH compared to the standard T2-STIR black-blood (standard T2-STIR) sequence and single-shot T2-STIR black-blood imaging (SSH T2-STIR) with breath-hold of basal, mid-ventricular, and apical slices. DL-SSH T2-STIR deep learning reconstructed single-shot T2-STIR black-blood imaging, BH breath-hold

• 

  Fig. 6. Cardiovascular magnetic resonance imaging examples in short-axis view from a 62-year-old participant after interventional treatment for occlusion of the left anterior descending coronary artery show subacute myocardial infarction. Myocardial edema is not clearly assessable in the standard T2-STIR images due to severe motion artifacts, due to the limited compliance of the patient. DL-SSH T2-STIR and T2-mapping clearly depict focal edema of the anterior and anterior-septal wall with corresponding LGE (yellow arrow), indicating subacute infarction. Standard T2-STIR Breath-hold multi-shot T2-weighted short tau inversion recovery sequence; DL-SSH T2-STIR deep learning reconstructed single-shot T2-STIR; LGE late gadolinium enhancement

• 

  Fig. 7. Cardiovascular magnetic resonance imaging examples in (A) a 62-year-old participant with purulent pericarditis, in (B) a 17-year participant with acute myocarditis, in (C) a 57-year participant with (sub) acute myocardial infarction and in (D) a 73-year participant with myocardial metastasis from renal cell carcinoma. Yellow arrows: Focal hyperenhancement. Standard T2-STIR Breath-hold multi-shot T2-weighted short tau inversion recovery sequence, DL-SSH T2-STIR deep learning reconstructed single-shot T2-STIR, LGE late gadolinium enhancement