Nowak, Sebastian: Development of AI-based methods for processing and quantitative analysis of radiological image data. - Bonn, 2023. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc:
author = {{Sebastian Nowak}},
title = {Development of AI-based methods for processing and quantitative analysis of radiological image data},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2023,
month = mar,

note = {Quantitative Image Analysis (QIA), that is, software-based extraction and analysis of numerically quantifiable features from medical imaging, has great potential to contribute to the progression of precision and personalized medicine. By utilizing objective, quantifiable features, biomarkers can be defined or predictive models can be developed that allow, e.g., the automatic detection of pathological alterations or the monitoring of disease progression or therapeutic success. To be practical in routine clinical care, QIA of tissues and organs should be automated and require minimal intervention by the radiologist. Artificial Intelligence (AI) methods and Deep Learning (DL) in particular have emerged as state-of-the-art image processing techniques in recent years, also for medical imaging.
This work features three AI-based pipelines for automated QIA. First, an end-to-end automated pipeline for quantification of muscle and adipose tissue (termed body composition analysis) is presented that includes automatic 2D slice extraction from 3D Computed Tomography (CT) scans, automatic tissue segmentation and quality control mechanisms to warn of potential invalid analysis. Then, a DL pipeline for automatic detection of liver cirrhosis in Magnetic Resonance Imaging (MRI) is demonstrated that features a method of explainable AI proposed to highlight image regions of importance. Finally, a pipeline for quantitative tissue assessment in MRI allowing also for monitoring of therapeutic success in patients with lip- and lymphedema is developed. This work includes a two-step anatomical landmark detection in combination with quality-assured tissue segmentation to create visualizations of tissue distribution in a standardized leg model.
The results of this work provide insights into the development of automated AI-based pipelines for use in clinical routine. Besides investigating methods for tissue segmentation, anatomical landmark and disease detection, it was also explored how combinations of those methods can be used in pipelines that overcome challenges of routine clinical data and thus minimize required effort of the radiologist as a prerequisite for potential use in clinical practice.},

url = {}

The following license files are associated with this item:

Namensnennung 4.0 International