Table 1.
Demographic data of the patients with acute MI subdivided by gender.
Fig 1.
Reformatted short axis images of the left ventricle.
Reformatted short axis images of the left ventricle at a slice thickness of 5 mm in a control (A) and in a patient with acute myocardial infarction (B) illustrating the free-hand regions-of-interest for texture analysis. Note the septal hypodensity indicating myocardial infarction (arrows).
Fig 2.
Principles of generating texture analysis features.
Principles of generating the histogram and the variables gray-level run-length matrix (GLRLM) and gray-level co-occurrence matrix (GLCM) from a given ROI. To construct the GLCM each pixel of the ROI is once compared with a pixel in a given distance and direction (0°, 45°, 90° or 135°). For each value pair the GLCM is increased by 1 in the respective column and row (in the given example a distance of 1 and direction of 90° were used; one 2–1 pair was found and 1 was added to the matrix accordingly). Runs of the same grey-level in a given direction (0°, 45°, 90° or 135°) are assessed to construct the GLRLM and used as the x-axis in the matrix, whereas the y-axis contains the assessed grey-levels. The example shows one run of three 2’s and a 1 is added to the matrix accordingly (a direction of 90° was used).
Fig 3.
Differences of texture analysis features in controls and patients with acute MI.
First- (histogram, left column), second- (GLCM, middle column), and third-level (GLRLM, right column) texture analysis features in a patient with acute myocardial infarction (upper row) and in a control (lower row). Note the additional peak at lower gray levels in patients with acute myocardial infarction indicating the proportion of voxels with a lower density and the divergent distribution of voxels in GLCM and GLRLM between controls and patients.
Table 2.
Intraclass correlation coefficients indicating the inter- and intrareader variability of all texture analysis features at each slice thickness.
Table 3.
Comparison of texture analysis features between controls and patients with acute MI for different slice thicknesses.
Values present the median and the interquartile range in parentheses.
Fig 4.
Boxplots of the three most distinguishing texture features.
Boxplots showing the three most distinguishing TA features between patients with acute MI and controls on images reformatted with a 5 mm slice thickness.
Fig 5.
ROC analysis showing the best model for predicting acute MI.
ROC analysis comparing the accuracy of the texture analysis features kurtosis (green; AUC: 0.78), correlation (orange; AUC: 0.81) and SRHGE (purple; AUC: 0.82) for predicting acute myocardial infarction. Combined analysis of kurtosis and SRHGE (red; mostly hidden behind the blue line, AUC: 0.9). Adding the parameter correlation to the first model added no benefit for the prediction of acute MI (blue; AUC: 0.9). Reference line in grey.