Automated cardiac segmentation pipeline and motion analysis in rodent models of pulmonary hypertension
File(s)
Author(s)
Niglas, Marili
Type
Thesis
Abstract
Right ventricular (RV) function is an independent predictor of survival in pulmonary arterial hypertension
(PAH) patients. Application of artificial intelligence (AI) techniques to cardiac magnetic resonance (CMR)
images improved ventricular phenotyping and survival projection by providing three-dimensional RV wall
motion patterns. While multiple AI-based cardiac image processing and analysis methods exist for clinical
application, the equivalent is lacking for rodent models, which offer longitudinal pathophysiological insight
into disease development, contributing extensively to treatment discovery. I proposed to develop a rat specific pipeline that enables automated bi-ventricular image and motion analysis. I hypothesised that the
pipeline will perform comparably to humans, and inform on the wall motion changes occurring throughout
pulmonary hypertension (PH) in rats.
I established a novel deep learning-based algorithm applied to rat CMR cine images producing bi-ventricular
segmentations, inclusive of RV annotations, for structural and functional analysis. The automated
segmentation approach enabled to produce a bi-ventricular segmentation in <1 second while maintaining
the accuracy comparable to human annotations. The segmentations generated were accurate across a wide
range of RV remodelling from control rats to adaptive and maladaptive stages in widely utilised PH-induced
rat models such as monocrotaline (MCT) and Sugen with chronic hypoxia (SuHx) male and female rats.
An atlas-fitting stage in the pipeline enabled to generate 3D cardiac ventricular models that complemented
and added additional insight into the bi-ventricular remodelling in PH animal models. The hypomobility
with PH development is in keeping with decreased functional global cardiac indices, and further enabled
distinction between adaptive and maladaptive rodents. The loss in longitudinal and radial RV motion at the
basal and mid-ventricular anterior regions, respectively, with altered LV longitudinal wall motion were most
distinct changes observed in PH rats. These findings mimic ventricular behaviour seen from PAH patients.
I further expanded the use of this methodology on therapeutic treatment follow-up in preclinical trials and
on more novel transgenic animal models to characterise bi-ventricular remodelling. The results showed that
the pipeline can be successfully applied for CMR follow-up therapeutic treatment assessment in rats treated
with chronic metabolic modulator GLP-1 receptor agonist liraglutide, and that the cardiac RV remodelling
was attenuated with the liraglutide in PH-induced rats. The BMPR2 transgenic animal model exhibited
reduced cardiac volumes at three-months timepoint without considerable differences in pulmonary
haemodynamics, yet with a tendency towards a lower RV pressure.
My PhD project contributed to the research field by providing a rapid and accurate CMR cine image
segmentation tool that, uniquely, generates RV segmentation annotations on rat images. Additional 3D
cardiac regional information can be used to characterise the PH-induced rodent cardiac remodelling process
in MCT and SuHx rats and response to treatment via longitudinal evaluation. This project has established
an automated computational platform for rodent CMR image processing, facilitating the identification and
extraction of disease progression markers that support translation between preclinical to clinical research.
(PAH) patients. Application of artificial intelligence (AI) techniques to cardiac magnetic resonance (CMR)
images improved ventricular phenotyping and survival projection by providing three-dimensional RV wall
motion patterns. While multiple AI-based cardiac image processing and analysis methods exist for clinical
application, the equivalent is lacking for rodent models, which offer longitudinal pathophysiological insight
into disease development, contributing extensively to treatment discovery. I proposed to develop a rat specific pipeline that enables automated bi-ventricular image and motion analysis. I hypothesised that the
pipeline will perform comparably to humans, and inform on the wall motion changes occurring throughout
pulmonary hypertension (PH) in rats.
I established a novel deep learning-based algorithm applied to rat CMR cine images producing bi-ventricular
segmentations, inclusive of RV annotations, for structural and functional analysis. The automated
segmentation approach enabled to produce a bi-ventricular segmentation in <1 second while maintaining
the accuracy comparable to human annotations. The segmentations generated were accurate across a wide
range of RV remodelling from control rats to adaptive and maladaptive stages in widely utilised PH-induced
rat models such as monocrotaline (MCT) and Sugen with chronic hypoxia (SuHx) male and female rats.
An atlas-fitting stage in the pipeline enabled to generate 3D cardiac ventricular models that complemented
and added additional insight into the bi-ventricular remodelling in PH animal models. The hypomobility
with PH development is in keeping with decreased functional global cardiac indices, and further enabled
distinction between adaptive and maladaptive rodents. The loss in longitudinal and radial RV motion at the
basal and mid-ventricular anterior regions, respectively, with altered LV longitudinal wall motion were most
distinct changes observed in PH rats. These findings mimic ventricular behaviour seen from PAH patients.
I further expanded the use of this methodology on therapeutic treatment follow-up in preclinical trials and
on more novel transgenic animal models to characterise bi-ventricular remodelling. The results showed that
the pipeline can be successfully applied for CMR follow-up therapeutic treatment assessment in rats treated
with chronic metabolic modulator GLP-1 receptor agonist liraglutide, and that the cardiac RV remodelling
was attenuated with the liraglutide in PH-induced rats. The BMPR2 transgenic animal model exhibited
reduced cardiac volumes at three-months timepoint without considerable differences in pulmonary
haemodynamics, yet with a tendency towards a lower RV pressure.
My PhD project contributed to the research field by providing a rapid and accurate CMR cine image
segmentation tool that, uniquely, generates RV segmentation annotations on rat images. Additional 3D
cardiac regional information can be used to characterise the PH-induced rodent cardiac remodelling process
in MCT and SuHx rats and response to treatment via longitudinal evaluation. This project has established
an automated computational platform for rodent CMR image processing, facilitating the identification and
extraction of disease progression markers that support translation between preclinical to clinical research.
Version
Open Access
Date Issued
2023-08-01
Date Awarded
01/12/2023
License URL
Advisor
Zhao, Lan
Bai, Wenjia
Baxan, Nicoleta
Sponsor
Engineering and Physical Sciences Research Council
Grant Number
EP/R513052/1
Publisher Department
National Heart & Lung Institute
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)