Cardiovascular Risk Assessment using Arterial Stiffness: A Deep Learning
Framework
Abstract
Background: Atherosclerotic cardiovascular disease (CVD) is severe and
early-stage detection is crucial. Elevated arterial stiffness observed
in childhood atherosclerosis is associated with CVD. Stiffness is an
efficient marker of CVD in hypertensives. Assessment of stiffness
includes waveform analysis and image-based techniques. Researchers
observed several challenges: real-time application, accuracy, operator
variability, image quality, scanning procedure, instrument variability
and deficiency of standardized procedure in the assessment. Methods: We
searched PubMed, Embase and Cochrane online library from inception up to
July 2020. Multiple articles on stiffness, pulse wave velocity,
assessment and deep learning (DL)-based methods were analysed. Above
all, a DL-based technique for assessment of stiffness from cine-loop is
proposed. The method includes region of interest (ROI) localisation in
multiple frames, segmentation of lumen and parameter estimation.
Results: Compared to conventional methods DL provide improved result in
lumen diameter and intima-media thickness (IMT) measurements. Using
convolutional neural network (CNN), IMT error was 0.08 mm. Further,
error using extreme learning machine-autoencoder was 5.79±34.42
\mum. Furthermore, Jaccard index and Dice similarity in
fully convolution neural network (FCN) manifested 0.94 and 0.97 for
lumen segmentation respectively. Conclusion: This paper focuses on the
association of stiffness and atherosclerosis leading to CVD. Success of
image-based stiffness estimation depends on the visibility and
orientation of arteries, operator experience, intensity variation,
shadowing, artefacts, and noise. Traditional methods include
transformations to compensate for these challenges. The success of
DL-based techniques in segmentation and localisation inspired
application in stiffness measurement. DL is used to estimate stiffness
from cine-loop.