Prognostic value of texture analysis from cardiac magnetic resonance imaging in patients with Takotsubo syndrome: a machine learning based proof-of-principle approach

Figures

• 

  Figure 1. Cardiac Magnetic Resonance Imaging and Texture Analysis. (A) Schematic depiction of acute takotsubo syndrome with characteristic ballooning of the left ventricle in two-chamber longitudinal view (top left image). The dotted line in green marks the plane for short axis cardiac magnetic resonance image acquisition (top right) at the point of maximal ventricular wall abnormality. After placement of a region-of-interest within the myocardium, a matrix of contained pixel intensities is generated (bottom left). Derived from inter-pixel relationships, various texture analysis features can be computed. Grey-Level co-occurrence matrices (GLCM) for 0° (bottom center) and 45° (bottom right) are shown. Exemplarily, the combination of two consecutive pixels with value ‘1’ is found twice in the signal intensity matrix and therefore marked as ‘2’ in the GLCM (0°) matrix at position row = 1 and colums = 1 (red box). Similarly, two consecutive pixels with value ‘1’ and ‘3’ are found once and therefore marked as ‘1’ in the GLCM (0°) matrix at position row = 1 and column = 3 (blue box). The diagonal runs of values ‘5’ and ‘2’ are found twice in the signal intensity matrix and therefore marked as…

• 

  Figure 2. Receiver operator characteristics for texture analysis on T2-weighted cardiac magnetic resonance images with five different machine learning classifiers of 5-year major adverse cardiovascular and cerebral events in patients with takotsubo syndrome. Of note, the NaïveBayes classifier shows the highest area-under-the-curve (0.88).

• 

  Figure 3. Box-Whisker plots of the ten selected texture analysis features. Dimension reduction yielded 10 TA features carrying prognostic information, which were all based on T2w images. Red: positive 5-year major adverse cardiovascular and cerebral events (MACCE), blue: negative 5-year MACCE. ** = p < .01.

• 

  Figure 4. Improving risk stratification in patients with takotsubo syndrome. Cardiac magnetic resonance imaging-derived texture analysis features, identified through machine learning algorithms, have a prognostic value in patients with takotsubo syndrome. Thus, these features might serve as novel imaging biomarkers for risk stratification in takotsubo syndrome.