Improved and Automated Detection of Papillary Muscle Infarction Using Joint Bright‐ and Black‐Blood Late Gadolinium Enhancement MRI

Figures

• 

  FIGURE 1. Automated PMI detection algorithm framework. (A) SPOT acquires 2D co‐registered black‐ and bright‐blood late gadolinium enhancement images in a single acquisition. For each slice position, eight single‐shot images (four black‐blood and four bright‐blood) are acquired and averaged. (B) Both black‐ and bright‐blood slices close to the apex and to the base are excluded according to the total number of slices of the scan. (C) Endocardial contours are segmented on the selected bright‐blood slices and (D) propagated onto corresponding co‐registered black‐blood slices. LV blood pool is then extracted, and a morphological erosion is computed. (E) A pixel intensity‐based criterion, μ, is used to determine if enhancement corresponding to a potential PMI is present. LV = left ventricular; PM = papillary muscle; PI = pixel intensity; PMI = papillary muscle infarction.

• 

  FIGURE 2. Papillary muscle (PM) ablation experiment on a sheep model. (A) Short‐axis echocardiographic images confirming the ablation catheter (green arrow) in position against the posterior PM (left, white arrow), and the lateral PM (right, white arrow). (B) 3D electroanatomical mapping showing the anatomical shell (left) and voltage map (right) of the left ventricle pre‐ablation (mapping catheter shown). Red, blue and yellow arrows represent the anteroseptal, anterolateral and posterior ablation regions, respectively. (C) Post‐ablation short‐axis MRI images, including the reference PSIR and the proposed SPOT images (black‐blood and colored fusion), displayed at three slice positions. The SPOT technique clearly improves the visualization of PM scars. (D) Histological analysis using Masson's trichrome staining and TTC staining, emphasizing the ablation sites. TTC = triphenyl‐tetrazolium chloride.

• 

  FIGURE 3. Examples of colored SPOT images (bottom) obtained by merging bright‐blood (second row) and black‐blood (third row) images, compared to reference PSIR imaging (top), in four patients with myocardial infarction. Yellow arrows indicate uncertain papillary muscle enhancements on PSIR imaging, clearly highlighted on SPOT imaging. PSIR = phase‐sensitive inversion recovery; yo = years‐old.

• 

  FIGURE 4. Comprehensive ventricular comparison between the reference PSIR images (top) and the proposed colored SPOT images (bottom), obtained by combining bright‐blood (second row) and black‐blood (third row) images, in a 64‐year‐old woman with myocardial infarction. PSIR = phase‐sensitive inversion recovery; yo = years‐old.

• 

  FIGURE 5. Six receiver operating characteristic (ROC) curves obtained by varying the number of morphological erosions applied to the left ventricular blood pool contour in the automated papillary muscle infarction detection algorithm. The ROC with five erosions achieved the highest AUC, with the optimal threshold (detection criterion, μ) determined to be 1.8 at the highest Youden index. AUC = area under the ROC curve.

• 

  FIGURE 6. Examples of automated PMI detection in four patients with myocardial infarction. Reference standard PSIR images are shown (left). In the SPOT bright‐blood images (second column), the endocardial contour (red) is segmented using a deep learning model. The contour is propagated to the corresponding co‐registered black‐blood image (third column), where a morphological erosion is applied to prevent residual endocardial enhancement (fourth column). The ratio of maximum to mean pixel intensity in the blood pool is then computed (right) and compared to the optimal detection threshold (1.8). PSIR = phase‐sensitive inversion recovery; LV = left ventricular; PM = papillary muscle; PI = pixel intensity; PMI = papillary muscle infarction; yo = years‐old.