Quantitative myocardial infarction on delayed enhancement MRI. Part I: Animal validation of an automated feature analysis and combined thresholding infarct sizing algorithm

• Published in: Journal of magnetic resonance imaging Vol. 23; no. 3; pp. 298 - 308
• Main Authors: Hsu, Li-Yueh, Natanzon, Alex, Kellman, Peter, Hirsch, Glenn A., Aletras, Anthony H., Arai, Andrew E.
• Format: Journal Article
• Language: English
• Published: Hoboken Wiley Subscription Services, Inc., A Wiley Company 01-03-2006
• Subjects: Algorithms; Animals; computer algorithm; contrast agent; Contrast Media; Dogs; expert system; gadolinium; Gadolinium DTPA; image processing; Image Processing, Computer-Assisted; magnetic resonance imaging; Magnetic Resonance Imaging - methods; myocardial infarction; Myocardial Infarction - pathology; Observer Variation; Tetrazolium Salts
• ISSN: 1053-1807; 1522-2586
• DOI: 10.1002/jmri.20496

Purpose
To develop a computer algorithm to measure myocardial infarct size in gadolinium‐enhanced magnetic resonance (MR) imaging and to validate this method using a canine histopathological reference.
Materials and Methods
Delayed enhancement MR was performed in 11 dogs with myocardial infarction (MI) determined by triphenyltetrazolium chloride (TTC). Infarct size on in vivo and ex vivo images was measured by a computer algorithm based on automated feature analysis and combined thresholding (FACT). For comparison, infarct size by human manual contouring and simple intensity thresholding (based on two standard deviation [2SD] and full width at half maximum [FWHM]) were studied.
Results
Both in vivo and ex vivo MR infarct size measured by the FACT algorithm correlated well with TTC (R = 0.95–0.97) and showed no significant bias on Bland Altman analysis (P = not significant). Despite similar correlations (R = 0.91–0.97), human manual contouring overestimated in vivo MR infarct size by 5.4% of the left ventricular (LV) area (equivalent to 55.1% of the MI area) vs. TTC (P < 0.001). Infarct size measured by simple intensity thresholdings was less accurate than the proposed algorithm (P < 0.001 and P = 0.007).
Conclusion
The FACT algorithm accurately measured MI size on delayed enhancement MR imaging in vivo and ex vivo. The FACT algorithm was also more accurate than human manual contouring and simple intensity thresholding approaches. J. Magn. Reson. Imaging 2006. Published 2006 Wiley‐Liss, Inc.