Categorisation of CT Reconstruction Kernels
Using Image Features Directly Extracted from Patient Scans
T.M. Camps (TU Delft - Mechanical Engineering)
M.C. Goorden – Mentor (TU Delft - Applied Sciences)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
CT is a versatile medical imaging method to diagnose and monitor patient diseases. However, varying patient characteristics and scan settings create challenges in maintaining consistent image quality, complicating image comparisons, especially across different sources. The reconstruction kernel in CT image reconstruction is a key parameter in the reconstruction process. It affects image characteristics such as sharpness, contrast, and noise. There is an urgent need for a method that effectively compares and categorises reconstruction kernels from different vendors using real patient scans. Therefore, this thesis focuses on extracting features from real patient images to facilitate kernel comparisons within and across manufacturers.
This research aims to create a machine learning (ML) method that categorises reconstruction kernels from various vendors into groups based on their sharpness. This categorisation relies on image features extracted directly from real patient scans with diverse scan parameters.
Two methods were explored using CT datasets from the National Lung Screening Trial (NLST) and the Lung Image Database Consortium (LIDC-IDRI). The first method uses noise features, specifically the standard deviation of homogeneous regions and the central frequency derived from the noise power spectrum. These features were used in a linear support vector machine (SVC_noise). The second method uses radiomic features extracted from selected homogeneous regions and is trained using a random forest classifier (RFC_radiomics).
Both models were evaluated using accuracy and ROC AUC. McNemar’s test was used to assess statistical differences. Due to the lack of ground truth, a subset of smooth and sharp kernels was used for training and validation, and remaining kernels were classified to establish a reference ground truth.
Both models performed strongly on 270 cases with 37 reconstruction kernels. The SVC_noise model achieved a ROC AUC of 0.97 with eight misclassifications, while the RFC_radiomics model achieved 0.96 with ten misclassifications. McNemar’s test showed no significant difference between the models. Only one discrepancy in ground truth assignment was observed for kernel “B50s”.
In conclusion, both models demonstrate strong and comparable performance in distinguishing kernel sharpness while being robust to variations in scan parameters and patient characteristics. However, results are preliminary and should be interpreted with caution when extending to other reconstruction kernels or imaging settings.