University of Twente Student Theses

Login

Conversion of CT kilovolt peak and reconstruction kernel using a 3D convolutional neural network for BoneMRI evaluation

Buijtenhuis, Bas René (2022) Conversion of CT kilovolt peak and reconstruction kernel using a 3D convolutional neural network for BoneMRI evaluation.

[img] PDF
11MB
Abstract:The combined use of CT and MRI allows for excellent diagnostic applications but requires an expensive and complex multimodal imaging workflow. MRIguidance has developed a deep learning approach to generate synthetic CT (BoneMRI) images using MRI data, which eliminates the need for a CT scan. Generation and evaluation of BoneMRI requires ground truth CT data. Evaluation metrics of BoneMRI should reflect its quality and should not depend on the quality or characteristics of the ground truth CT images. However, it was found that the kilovolt peak (kVp) and reconstruction kernel have an influence on the BoneMRI evaluation metrics. The goal of this research was therefore to standardize CT data characteristics in terms of kVp and reconstruction kernel, by developing image- to-image conversion methods. Two methods were developed to convert CT images from a low (100) kVp or high (140) kVp to a target kVp of 120. The first method was a piecewise linear fit, while the second method used a convolutional neural network (CNN). A separate CNN was trained to convert images from a soft reconstruction kernel to a sharp reconstruction kernel. The kVp conversion methods were used to convert the CT data being used for BoneMRI evaluation. The evaluation metrics of BoneMRI predictions were calculated with- and without using the developed conversion methods. The study revealed that the conversion of CT images from one kVp to another kVp is possible using a piecewise linear fit or a CNN if paired data is available. The evaluation metrics of BoneMRI improved by applying the kVp conversion methods on the ground truth CT data. A CNN can also be used to convert CT images that were reconstructed with a particular reconstruction kernel to another reconstruction kernel. The conversion methods did not generalize to CT-data that was different than the CT-data that was present within the training data of the conversion methods.
Item Type:Essay (Master)
Faculty:TNW: Science and Technology
Subject:50 technical science in general
Programme:Biomedical Engineering MSc (66226)
Link to this item:https://purl.utwente.nl/essays/93113
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page