Publicação
Comparison of T1-maps and late gadolinium enhancement images in the detection of Myocardial Fibrosis in Hypertrophic Cardiomyopathy
| Resumo: | Hypertrophic Cardiomyopathy (HCM) is characterized as an abnormal and heterogeneous thickening of the Left Ventricle (LV) wall. HCM is the leading cause of sudden cardiac death in children and young people, with an estimated prevalence of 1:500 in the general population. Myocardial fibrosis is the key histopathological hallmark in HCM and is presented in different patterns: interstitial diffuse fibrosis which, if not treated, evolves to replacement fibrosis. Cardiac Magnetic Resonance (CMR) imaging has been used for the detection and quantification of myocardial fibrosis. The Late Gadolinium Enhancement (LGE) technique is the primary tool for non-invasive tissue characterization, particularly for replacement fibrosis. Conversely, T1 mapping is commonly used for the detection of diffuse interstitial fibrosis, frequently missed using LGE. The clear disadvantage of LGE relies on the need to inject contrast agents that, despite being considered safe, may accumulate in the body for years and potentially cause nephrogenic systemic fibrosis in end-stage chronic kidney disease patients. The capability of native T1 mapping identifying not only diffuse interstitial but also replacement fibrosis would play a pivotal role in HCM diagnosis. The potential of native T1 mapping for a cheaper and non-contrast HCM assessment needs to be further studied. A database of 15 HCM patients, without and with fibrosis, was acquired at Hospital da Luz, Lisboa. In this project, (1) an extensive image preprocessing pipeline was applied to aim for the best possible spatial alignment of the myocardium between the two modalities (native T1 mapping and LGE); (2) the mean native T1 values of individuals without and with the presence of scarred tissue were examined; (3) a pixel-by-pixel analysis was performed to investigate if there is a correlation between fibrotic tissue in LGE and hyperintense regions in native T1 mapping; (4) a Texture Analysis (TA) was performed to study if texture information of native T1 mapping could provide differential diagnosis or prognostic information beyond mean T1 values. The first step was the most longstanding and challenging process. The registration of T1 and LGE images is difficult due to the different intensity profiles. The registration of the myocardial masks using a model with rigid, affine, and free-form deformation transformations revealed to be the best methodology. Mean native T1 values were not increased in patients with scarred tissue. Regarding the third aim, no clear intensity correlation between techniques was observed, which suggests the need for the TA. Seven features (in a total of 350) were selected to distinguish between cardiac segments without and with fibrotic tissue using a ML (Machine Learning) algorithm that finds the features that most contribute to distinguish the two groups. Four first-order features distinguish the cohorts due to the presence of scarred tissue - hyperintense zones - and three texture features suggest that the fibrotic remodeling in the myocardium of HCM patients might be associated with a more heterogeneous tissue texture. A Receiver Operating Characteristics (ROC) analysis was performed and revealed that the Cluster Prominence is the feature that best distinguishes sections without and with fibrotic tissue (accuracy of 70%) but with low sensitivity (65%) and low specifity (64%). A model with the 90th Percentile feature revealed an accuracy of 64%, sensitivity of 71% and specificity of 57%. Studying the Variance feature, the achieved accuracy was 63%, with 66% of sensitivity and 60% of specificity. The remaining features yielded lower accuracy values than the ones previously mentioned, but all of them higher than 50%. The low sensitivity and specificity of the best three models suggest that analysing these values considering these features may help cardiologists to identify focal fibrosis regions and avoid contrast injection methods but may not provide an accurate diagnosis of the presence of fibrotic tissue alone. Further research on the correlation of native T1 mapping and LGE cardiac images is highly recommended to develop a contrast-agent-free technology to replace LGE. |
|---|---|
| Autores principais: | Baleia, Cláudia Neves Silvestre |
| Assunto: | Fibrose do miocárdio Realce tardio com gadolínio Mapeamento T1 Análise de textura Teses de mestrado - 2022 |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | Hypertrophic Cardiomyopathy (HCM) is characterized as an abnormal and heterogeneous thickening of the Left Ventricle (LV) wall. HCM is the leading cause of sudden cardiac death in children and young people, with an estimated prevalence of 1:500 in the general population. Myocardial fibrosis is the key histopathological hallmark in HCM and is presented in different patterns: interstitial diffuse fibrosis which, if not treated, evolves to replacement fibrosis. Cardiac Magnetic Resonance (CMR) imaging has been used for the detection and quantification of myocardial fibrosis. The Late Gadolinium Enhancement (LGE) technique is the primary tool for non-invasive tissue characterization, particularly for replacement fibrosis. Conversely, T1 mapping is commonly used for the detection of diffuse interstitial fibrosis, frequently missed using LGE. The clear disadvantage of LGE relies on the need to inject contrast agents that, despite being considered safe, may accumulate in the body for years and potentially cause nephrogenic systemic fibrosis in end-stage chronic kidney disease patients. The capability of native T1 mapping identifying not only diffuse interstitial but also replacement fibrosis would play a pivotal role in HCM diagnosis. The potential of native T1 mapping for a cheaper and non-contrast HCM assessment needs to be further studied. A database of 15 HCM patients, without and with fibrosis, was acquired at Hospital da Luz, Lisboa. In this project, (1) an extensive image preprocessing pipeline was applied to aim for the best possible spatial alignment of the myocardium between the two modalities (native T1 mapping and LGE); (2) the mean native T1 values of individuals without and with the presence of scarred tissue were examined; (3) a pixel-by-pixel analysis was performed to investigate if there is a correlation between fibrotic tissue in LGE and hyperintense regions in native T1 mapping; (4) a Texture Analysis (TA) was performed to study if texture information of native T1 mapping could provide differential diagnosis or prognostic information beyond mean T1 values. The first step was the most longstanding and challenging process. The registration of T1 and LGE images is difficult due to the different intensity profiles. The registration of the myocardial masks using a model with rigid, affine, and free-form deformation transformations revealed to be the best methodology. Mean native T1 values were not increased in patients with scarred tissue. Regarding the third aim, no clear intensity correlation between techniques was observed, which suggests the need for the TA. Seven features (in a total of 350) were selected to distinguish between cardiac segments without and with fibrotic tissue using a ML (Machine Learning) algorithm that finds the features that most contribute to distinguish the two groups. Four first-order features distinguish the cohorts due to the presence of scarred tissue - hyperintense zones - and three texture features suggest that the fibrotic remodeling in the myocardium of HCM patients might be associated with a more heterogeneous tissue texture. A Receiver Operating Characteristics (ROC) analysis was performed and revealed that the Cluster Prominence is the feature that best distinguishes sections without and with fibrotic tissue (accuracy of 70%) but with low sensitivity (65%) and low specifity (64%). A model with the 90th Percentile feature revealed an accuracy of 64%, sensitivity of 71% and specificity of 57%. Studying the Variance feature, the achieved accuracy was 63%, with 66% of sensitivity and 60% of specificity. The remaining features yielded lower accuracy values than the ones previously mentioned, but all of them higher than 50%. The low sensitivity and specificity of the best three models suggest that analysing these values considering these features may help cardiologists to identify focal fibrosis regions and avoid contrast injection methods but may not provide an accurate diagnosis of the presence of fibrotic tissue alone. Further research on the correlation of native T1 mapping and LGE cardiac images is highly recommended to develop a contrast-agent-free technology to replace LGE. |
|---|