Tomographic back-projection of either sparse or low-quality projection views, based on convolutional neural networks (CNN)

Payal JAIN

Master's thesis

Abstract:

This thesis delves into the integration of classical and modern approaches in tomographic back-projection, with a particular emphasis on challenges presented by sparse or low-quality projection views. The research is dedicated to advancing reconstruction methods by combining Convolutional Neural Networks (CNNs) with classical algorithms, specifically the Feldkamp-Davis-Kress (FDK) and Simultaneous Iterative Reconstruction Technique (SIRT) algorithms implemented using the ASTRA Toolbox. Evaluation of the methodologies relies on key performance metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), and Structural Similarity Index (SSIM). The experimental results underscore the effectiveness of CNNs in overcoming challenges associated with sparse or low-quality projection views, providing valuable insights for the enhancement of tomographic reconstruction methodologies.

Abstract:

This thesis delves into the integration of classical and modern approaches in tomographic back-projection, with a particular emphasis on challenges presented by sparse or low-quality projection views. The research is dedicated to advancing reconstruction methods by combining Convolutional Neural Networks (CNNs) with classical algorithms, specifically the Feldkamp-Davis-Kress (FDK) and Simultaneous Iterative Reconstruction Technique (SIRT) algorithms implemented using the ASTRA Toolbox. Evaluation of the methodologies relies on key performance metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), and Structural Similarity Index (SSIM). The experimental results underscore the effectiveness of CNNs in overcoming challenges associated with sparse or low-quality projection views, providing valuable insights for the enhancement of tomographic reconstruction methodologies.

Keywords

Tomography, reconstruction, neural network, CNN, FDK, performance matrics, astra toolbox

Language used: English

Date on which the thesis was submitted / produced: 9. 2. 2024

Thesis defence

• Supervisor: prof. Dr. Simon Zabler

UNIVERSITY OF SOUTH BOHEMIA IN ČESKÉ BUDĚJOVICE

Faculty of Science

Master programme / field:
Artificial Intelligence and Data Science / Artificial Intelligence and Data Science