Texture Analysis of Late Gadolinium Enhanced Cardiac Magnetic Resonance Images for Characterizing Myocardial Fibrosis and Infarction

A third of the United States population is affected by cardiomyopathies. Impairment of the heart muscle, the myocardium, puts the patient's health at risk and could ultimately lead to death. Ischemic cardiomyopathies result from lack of blood (ischemia) reaching the myocardium from blocked coronary arteries. Non-ischemic cardiomyopathies are diseases from other etiology than ischemia. Often collagen fibers infiltrate the heart (fibrosis), as a means to maintain its shape and function in the presence of disease that affects the myocardial cellular structure. This necessary phenomenon ultimately becomes maladaptive and results in the heart's impairment. Part of the heart's involvement in disease can be assessed through the analysis of myocardial fibrosis. Cardiomyopathy diagnosis involves the investigation of the presence of myocardial fibrosis, either infiltrative, defined as the increased presence of collagen protein in the extracellular space, or replacement fibrosis, when collagen fibers progressively replace diseased cardiomyocytes. The infiltrative fibrosis is believed to be reversible in some instances and consequently, myocardial fibrosis analysis has decisional impact on the interventional procedure that would benefit the health of the patient. The heart contracts and relaxes as it pumps blood to the rest of the body, an action directly impaired by myocardial damage. Any myocardial involvement should be assessed by the clinician to identify the severity of the myocardial damage, establish a prognosis and plan therapeutic intervention. Different diagnostic tests are required to image the myocardium and help the clinician in the diagnostic process. Cardiac magnetic resonance (CMR) imaging has emerged as a high resolution imaging modality that offers precise structural analysis of the heart. Among the different imaging sequences available with CMR, late gadolinium enhancement (LGE) shows the myocardium and enhances any impairments that may exist with the use of a contrast agent. It is a T1-weighted image with extracellular contrast agent (CA) administration. Increased signal intensity in the infarct scar is created from the CA dynamics. LGE CMR imaging offers information on the scar size and its location. The clinician can estimate the severity of the disease and establish prognosis with LGE CMR images. In ischemic cardiomyopathy, it is important to establish the occurrence of the infarction and know the age of the infarct to plan surgical intervention. Differentiation of acute from chronic MI is therefore important in the diagnostic process. In LGE CMR the level of signal intensity or the size of infarction are both similar in acute or in chronic MI. It has therefore been challenging to distinguish acute MI from chronic MI scars with LGE CMR images alone. The aim of this thesis was to investigate texture analysis of LGE CMR images to determine if acute MI could be distinguished from chronic MI and to detect increased presence of diffuse myocardial fibrosis in the myocardium. The first study was performed to investigate if texture analysis of LGE CMR images could detect variations in the presence of diffuse myocardial fibrosis and if the underlying myocardial structure could be related to the texture measures. Collagen content increased with aging and we used an animal model of young versus old rat. An ex-vivo animal model was necessary to allow for higher image resolution in LGE CMR images and to perform validation of our texture measures with histology images. Twenty four animals were scanned for LGE CMR images and texture analysis was applied to the heart images. Histology slices were stained with picrosirius red and collagen fibers were isolated based on their color content. LGE CMR quantification was compared to histological slices of the heart stained with the picrosirius red method. Texture analysis of LGE CMR images was also compared to the original LGE CMR image quantification and to histology. Texture analysis was done by creating contrast texture maps extracted from Haralick's gray level co-occurrence matrix (GLCM). Regions of complex signal intensity combination were enhanced in LGE CMR images and in contrast texture maps. Regression analysis was performed to assess the level of agreement between the different analysis methods. We found that LGE CMR images could assess the different levels of collagen content in the different aged animal model, and that moreover texture analysis enhanced those differences. The location of enhancement from texture analysis images corresponded to location of increased collagen content in the old compared to the young rat hearts. Histological validation was shown for texture analysis applied to LGE CMR images to assess myocardial fibrosis. Our second study aimed at discriminating acute versus chronic MI from LGE CMR patient images alone through the use of texture analysis. Twenty two patients who had LGE CMR images were included in our study (12 acute and 12 chronic MI). Regional segmentation was performed and texture features were compared in those regions between both groups of patient. Texture analysis resulted in significantly different values between the two groups. More specifically the peri-infarct zone had the most number of discriminative features compared to the remote myocardium which had none and to the infarct core where only the mean features was significantly different. The border zone has been shown to be composed of healthy cardiomyocytes intermingled with the scar's collagen fibers. Our study indicates this region might exhibit structural differences in the myocardium in acute from chronic MI patients that texture analysis of LGE CMR images can detect. Characterization of myocardial collagen content is important while clinicians analyze the state of the patient since it influences the course of action required to treat cardiomyopathies. LGE CMR images have been thoroughly used and validated to characterize focal myocardial scar, however it was limited in characterizing the age of infarction or quantifying diffuse collagen content. We have shown texture analysis of LGE CMR images alone can differentiate and even classify, acute from chronic MI patients, which was not previously possible. Characterization of myocardial infarction according to age will prove important in planning therapeutic interventions in clinical practice. Moreover, we have established texture analysis as a means to characterize the myocardium and detect variation in fibrosis content from high resolution LGE CMR images with histology validation. To our knowledge, such a relation between texture analysis of LGE CMR images and the underlying myocardial structure had not been done previously. Improvements could be done to our method, as we can increase the number of texture features that were analyzed from the GLCM, include other texture analysis methods such as the run-length matrix, and apply our method to other CMR imaging sequences such as T1 mapping. Adapting the GLCM to the heart could also be investigated, such as considering circular GLCM computation to consider the round shape of the myocardium in the short axis LGE CMR image slices. Directional GLCM could also be computed individually and analyzed for any myocardial or collagen fiber orientation indication. Further analysis is also required to establish if texture analysis could differentiate the age of MI in the same individual through a follow-up study. The measures of texture analysis from LGE CMR images obtained through different CMR scanners remains to be investigated as well. Knowing the age of infarct and evaluating the presence of diffuse myocardial fibrosis will help the clinician plan therapeutic interventions and establish a prognosis for the patient.