Texture-based probability mapping for automatic scar assessment in late gadolinium-enhanced cardiovascular magnetic resonance images

Figures

• 

  Fig. 1. Varieties of signal intensity of the matrices. Fig. 1 shows a LGE-MRI with a myocardial ischemic scar (bright pixels delineated by a blue line). P1: Core. P2: Border zone. P3-P4: Normal myocardium, but still, there is a variety in signal intensity. Fig. 2.

• 

  Fig. 2. Assessment by the TPM-method a) Original LGE-CMRI with Corresponding TPM color map (b). Shades of blue indicate normal myocardium, and green, yellow and orange indicate scarred myocardium with higher probability as the color gets more intense. We can appreciate the smooth transition between normal and scarred myocardium which is thought to make more sense physiologically compared with the crisp segmentation by the hand of the cardiologist. c) TPM with a cutoff threshold of 0.328. All pixels above that threshold are red, and pixels below are green.

• 

  Fig. 3. Pixel distribution The figure displays how the probability distributes for all pixels of the 52 patients analyzed in the present study. Since the manual delineation is based on visual interpretation of SI, and the TPM is based on replicating this from texture information, the recognition accuracy and the probabilities will not reach higher than 0.75. The probability threshold between scarred and non-scarred pixels comparing the TPM method with the manual demarcation was found to be 0.328. Pixels with a probability value below this limit were defined as non-scar (green), whereas pixels with a probability value above this limit are considered to represent scarred tissue (red).

• 

  Fig. 4. Cross-sectional images of infarcted myocardium visualized by different methods a) Original cross sectional LGE-CMRI with Corresponding TPM color map without cutoff (b). Figure c-f illustrates the different scar delineation by using the four methods. Green and red colors indicate normal and scarred myocardium, respectively. The Scar is defined as pixels with probability mapping-values ≥ 0.328 (red), while pixels below this threshold are classified as normal myocardium (green).

• 

  Fig. 5. a-c) Bland Altman plot of the TPM method compared with manual segmentation, FWHM and the 5 SD-technique. The Bland Altmann plots a-c show the agreement between the TPM-method and the SI-based techniques. The central line indicates the median difference, while the dotted lines represent the limits of agreement (median ± 1.96 x SD). With higher average scar size, the TPM method seems to estimate a higher scar size then FWHM. There is no significant difference between scar sizes estimated by the TPM method and the 5 SD-technique.

• 

  Fig. 6. Gives a summary of the creation of the TPM color map and highlights the selected probability range used to assess scar.