4.3 Challenges in image-based methods
Thus, conventional segmentation methods are either semi-automatic or
automatic. Human intervention is required at many stages, for example,
ROI selection, seed point and/or contour initialization, and distance
evaluation [122]. This leads to additional latency and is often
error prone. Further, the curvature of the artery and its orientation
impinge on segmentation [122]. Most segmentation algorithms require
arteries to be recorded horizontally. Spatial transformations are
applied to improve orientation before segmentation [122].
Furthermore, the presence of plaque, plaque irregularity and mimicking
arterial structures such as jugular vein complicates the process.
Moreover, scanner variability, operator experience, angle of incidence
(of ultrasound probe), blood back-scattering, shadowing and ultrasound
artefacts add to the conundrum [123]. Application of machine
learning and deep learning algorithms provided a competitive edge in
this scenario.