Information
- Publication Type: Master Thesis
- Workgroup(s)/Project(s): not specified
- Date: 2023
- Open Access: yes
- First Supervisor: Renata Raidou
- Pages: 136
- Keywords: Visual Analytics, Big Data, Radiomics, Genomics, Cancer
Abstract
Radiogenomics refers to the combined study of imaging-derived features, called radiomics and gene sequencing data, called genomics. Challenges in the analysis of radiogenomic data include the size, heterogeneity, and complexity of the datasets. These challenges make the analysis of the available information space tedious for cancer experts and hinder the exploration and sensemaking of patient information. This is further hampered when additional clinical information needs to be included in the analyses. Visual Analytics (VA) combines automated analysis techniques, such as machine learning or statistics, together with interactive visual interfaces. It allows users to gain insights into complex data and make effective decisions. In the context of radiogenomics analysis with respect to clinical data, VA approaches offer promising directions in tumor profiling. However, VA approaches that bridge radiogenomic and clinical data in an interactive and flexible visual framework have not been investigated before. In this work, we enable the integrated exploration and analysis of radiogenomic data and clinical information for knowledge discovery and hypothesis assessment in a large cohort of prostate cancer patients. We handle missingness in the data through imputation techniques and apply unsupervised machine learning for the dimensionality reduction and clustering of the data to facilitate data handling and visualization. As a result, we present an interactive visual interface for two target audiences: cancer experts and biomedical data scientists. Our framework enables cancer experts to gain insights into the data by revealing new patterns or correlations in the datasets. It allows them to interactively assess and refine any hypothesis in mind for the underlying datasets. For biomedical data scientists, our framework offers the possibility to understand the analysis components and interactively explore their impact on the outcome. We evaluate the unsupervised machine learning models through similarity measures such as the silhouette coefficient. To assess the usability of the framework, we perform usage scenarios that we confirm by our cancer experts. The feedback from our domain experts reveals that our framework is a suitable and flexible technique to gain insights into large and heterogenous radiogenomic data with respect to clinical data. It promotes knowledge discovery as well as hypothesis creation, assessment, and refinement. Interacting with the different visualization and analysis components enhances the understanding of the data and the resulting visual representations. Our approach incorporates the integration of interactive visualization and automated analysis components. It supports our collaborating domain experts at the Medical University of Vienna to obtain new insights into their data, while investigating hypotheses at hand.
Additional Files and Images
Weblinks
BibTeX
@mastersthesis{el-sherbiny-2023-vat,
title = "Visual Analytics to Support Correlative Exploration and
Sensemaking in Radiogenomics Analysis",
author = "Sarah El-Sherbiny",
year = "2023",
abstract = "Radiogenomics refers to the combined study of
imaging-derived features, called radiomics and gene
sequencing data, called genomics. Challenges in the analysis
of radiogenomic data include the size, heterogeneity, and
complexity of the datasets. These challenges make the
analysis of the available information space tedious for
cancer experts and hinder the exploration and sensemaking of
patient information. This is further hampered when
additional clinical information needs to be included in the
analyses. Visual Analytics (VA) combines automated analysis
techniques, such as machine learning or statistics, together
with interactive visual interfaces. It allows users to gain
insights into complex data and make effective decisions. In
the context of radiogenomics analysis with respect to
clinical data, VA approaches offer promising directions in
tumor profiling. However, VA approaches that bridge
radiogenomic and clinical data in an interactive and
flexible visual framework have not been investigated before.
In this work, we enable the integrated exploration and
analysis of radiogenomic data and clinical information for
knowledge discovery and hypothesis assessment in a large
cohort of prostate cancer patients. We handle missingness in
the data through imputation techniques and apply
unsupervised machine learning for the dimensionality
reduction and clustering of the data to facilitate data
handling and visualization. As a result, we present an
interactive visual interface for two target audiences:
cancer experts and biomedical data scientists. Our framework
enables cancer experts to gain insights into the data by
revealing new patterns or correlations in the datasets. It
allows them to interactively assess and refine any
hypothesis in mind for the underlying datasets. For
biomedical data scientists, our framework offers the
possibility to understand the analysis components and
interactively explore their impact on the outcome. We
evaluate the unsupervised machine learning models through
similarity measures such as the silhouette coefficient. To
assess the usability of the framework, we perform usage
scenarios that we confirm by our cancer experts. The
feedback from our domain experts reveals that our framework
is a suitable and flexible technique to gain insights into
large and heterogenous radiogenomic data with respect to
clinical data. It promotes knowledge discovery as well as
hypothesis creation, assessment, and refinement. Interacting
with the different visualization and analysis components
enhances the understanding of the data and the resulting
visual representations. Our approach incorporates the
integration of interactive visualization and automated
analysis components. It supports our collaborating domain
experts at the Medical University of Vienna to obtain new
insights into their data, while investigating hypotheses at
hand.",
pages = "136",
address = "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
school = "Research Unit of Computer Graphics, Institute of Visual
Computing and Human-Centered Technology, Faculty of
Informatics, TU Wien",
keywords = "Visual Analytics, Big Data, Radiomics, Genomics, Cancer",
URL = "https://www.cg.tuwien.ac.at/research/publications/2023/el-sherbiny-2023-vat/",
}
thesis