AI-Based Quality Assessment of Clinical Thorax CT Images
The quality of computer tomography (CT) images can be affected by several technical parameters and external factors including tube settings, patient motion, implants or anatomical structures. If not checked immediately, the patient might need to visit the radiologist again for an additional scan. In order to avoid this scenario, it would be advantageous to automatically analyze, report and document the quality of a CT dataset directly following its acquisition.
Various traditional image quality metrics, such as image noise or the modulation transfer function, among others, might not be suitable for the complexity of clinical CT images. The scope of this thesis is to investigate if AI/ML techniques are suitable to tackle this problem.
A thorax (chest) CT dataset of various patients having different pathologies or diseases was labelled by different radiologists. Based on this dataset, one or more techniques for analyzing the image quality shall be selected, implemented and compared. The final goal is to provide a quality score for each patient CT record.
The following aspects should be considered within the thesis:
- Literature survey of state of the art considering AI/ML techniques, as well as their application for the given purpose.
- Review, analysis and preparation of the provided CT data.
- Selection and implementation of the AI/ML methods and their application to the given CT dataset.
- Critical results analysis.
The task can be adjusted to the interests and experience of the student. The results of the thesis shall be summarized as a 4-8 page scientific paper.
