Combining Static and Dynamic Features to Improve Longitudinal Image Retrieval for Alzheimer's Disease
Date Issued
2022-09-29
Author(s)
Trojachanec Dineva, Katarina
Abstract
The aim of this paper is to enhance medical case retrieval for Alzheimer’s disease on the basis of the domain knowledge. We approached the problem in a longitudinal manner, and we represented the medical cases by using
different kind of information extracted from Magnetic Resonance Images (MRI)
aiming to improve the semantic relevance, precision and efficiency of the retrieval. More particularly, we evaluated the combination of the static, dynamic
features and the index reflecting the spatial pattern of abnormality (SPARE-AD)
for representing the longitudinal images. According to the obtained results, the
combination of the static features representing the volumetric measures along
with the cortical thickness measures of the brain structures at the later time point/s
together with the dynamic features such as percent change with respect to the
value obtained from the linear fit at baseline and symmetrized percent change of
the volumetric measures, as well as the index of abnormality provided the best
overall retrieval results. The dimensionality of the feature vector was 31-33 features in most of the cases which is significantly lower than in the case of the
traditional approach (thousands features in the cases when the whole brain is considered). The approach based on a combination of different kinds of features extracted from the longitudinal data, suggested in this paper, corresponds directly
to the nature of the application domain and provides powerful results, yet effective and efficient way for MRI retrieval for AD.
different kind of information extracted from Magnetic Resonance Images (MRI)
aiming to improve the semantic relevance, precision and efficiency of the retrieval. More particularly, we evaluated the combination of the static, dynamic
features and the index reflecting the spatial pattern of abnormality (SPARE-AD)
for representing the longitudinal images. According to the obtained results, the
combination of the static features representing the volumetric measures along
with the cortical thickness measures of the brain structures at the later time point/s
together with the dynamic features such as percent change with respect to the
value obtained from the linear fit at baseline and symmetrized percent change of
the volumetric measures, as well as the index of abnormality provided the best
overall retrieval results. The dimensionality of the feature vector was 31-33 features in most of the cases which is significantly lower than in the case of the
traditional approach (thousands features in the cases when the whole brain is considered). The approach based on a combination of different kinds of features extracted from the longitudinal data, suggested in this paper, corresponds directly
to the nature of the application domain and provides powerful results, yet effective and efficient way for MRI retrieval for AD.
Subjects
File(s)![Thumbnail Image]()
Loading...
Name
2. ICT2022_StaticAndDynamicF_LongCBIR_AD.pdf
Size
1.06 MB
Format
Adobe PDF
Checksum
(MD5):78f3b0d620c944fb73eb13cc694d6e16