an Automated Feature Analysis and Combined Thresholding Infarct Sizing Algorithm

Purpose: To compare global and regional myocardial in-farction (MI) measurements on clinical gadolinium-en-hanced magnetic resonance (MR) images using human manual contouring and a computer algorithm previously validated by histopathology, and to study the degree to which visual assessment and human contouring of infarct extent agreed with the computer algorithm. Materials and Methods: Infarct size in 20 patients was measured by human manual contouring and with an au-tomated feature analysis and combined thresholding (FACT) computer algorithm. Short-axis slices were divided into myocardial sectors for regional analysis. Extent of in-farction was also graded visually by consensus of expert readers and compared to human and computer contouring. Results: Despite good correlations (R 0.93–0.95) be-tween human contouring and the FACT algorithm, human contouring overestimated infarct size by 3.8 % of the left ventricle (23.8 % of the MI) area (P 0.001). Human con-touring also overestimated the circumferential extent, transmural extent, and extent of infarction within a sector by 7.1%, 18.2%, and 27.9%, respectively (all P 0.001). Both consensus reading and human contouring overesti-mated infarct grades compared with the FACT algorithm (P 0.002 and P 0.001). Conclusion: Clinically relevant overestimation of MI can occur in visual interpretation and in human manual con-touring, particularly with respect to extent of infarction on a regional basis. Key Words: myocardial infarction; myocardial viability; magnetic resonance imaging; computer aided diagnosis; contrast agent; gadoliniu