UMS-Net++: Modified SwinTransformer for Cardiac MRI Segmentation and Classification

Cardiac disease detection using Magnetic Resonance Imaging (MRI) inputs is encouraged
using intelligent computer-aided visual processing techniques in smart health. The pixeldependent
segmentation
and
classification
are
automated
for
disease
detection
and
diagnosis

in

its early stages using conventional image processing. This article introduces a UnitModified
SwinTransformer
Network
(UMSN)
for
thwarting
pixel
differentiation
issues
in

cardiac

MRI analysis. The proposed network is different from the conventional
SwinTransformer Network by grouping similar dimension pixels from 2×2 to 16×16
segmentation. In this grouping the analysis layer is trained using unanimous feature
extractable pixels and the end-of-pixel region is the segment boundary detected. In this
detection process, the coinciding pixel boundaries with similar features are used for training
the fundamental segment of the consecutive boundaries. Therefore, the disparity between
grouped and independent boundary segments is identified as a false rate. This identified unit
is used for independent training across various labeled classifications from the training
inputs. Therefore, the disparity between the boundaries or segment pixels is used for
identifying flaws in the input MRI.