Files and links (1)
PDFOpen Access (License Unspecified), Open Access
Thesis
Automated evaluation pipeline of CT-CBCT deformable image registration
Master of Science (M.S.), Drexel University
Dec 2024
DOI:
https://doi.org/10.17918/00010841
Abstract
Cone Beam Computed Tomography (CBCT) images are low quality and often suffer from artifacts and noise which makes it hard to use them for locating tumors. CT images are high quality images with significantly less artifacts and noise, but are usually at a higher cost than the CBCT. Improving the quality of the medical images using AI methods has been growing recently. Assessing the quality of such synthetic data is also important to improve the quality of the models. High level of importance are given to minor details when it comes to medical images which usually involves multiple stages of processing the data. Automating these stages and reducing the amount of human intervention plays a vital role in speeding up the AI models prototyping. This thesis focus on building an automated evaluation pipeline for synthetic CT images produced using Cycle-GAN with the flexibility to tune every single parameters during evaluation. The pipeline uses different features like segments and distance maps to perform deformable registration and calculate metrics like DICE Similarity Coefficient and Mean Fiducial Separation to assess the quality of the generated synthetic images. The pipeline utilizes open- source softwares like plastimatch and TotalSegemntator to aid the processes in the evaluation pipeline and produced an average DSC of 92% and and Fiducial separation of 8mm. The proposed method can be used for evaluating synthetic medical images with the flexibility to use any pre-trained segmentation model out of the box.
Metrics
403 File views/ downloads
19 Record Views
Details
- Title
- Automated evaluation pipeline of CT-CBCT deformable image registration
- Creators
- Santhosh Vadivel
- Contributors
- Nagarajan Kandasamy (Advisor)
- Awarding Institution
- Drexel University
- Degree Awarded
- Master of Science (M.S.)
- Publisher
- Drexel University; Philadelphia, Pennsylvania
- Number of pages
- v, 36 pages
- Resource Type
- Thesis
- Language
- English
- Academic Unit
- College of Engineering (1970-2026); Electrical (and Computer) Engineering [Historical]; Drexel University
- Other Identifier
- 991022020017004721