Computational Toolkit for CT Acquisition and Reconstruction

Normalizing the Impact of CT Acquisition and Reconstruction on Quantitative Image Features

Doctor typing on touchscreen
By Medical & Imaging Informatics (MII)
1 min read

Quantitative image features (QIFs), such as radiomic and deep features derived from Computed Tomography (CT) scans, hold immense potential for improving disease detection, diagnosis, and treatment assessment. However, the consistency and reproducibility of these QIFs are significantly impacted by variations in CT acquisition and reconstruction parameters, such as radiation dose, slice thickness, and reconstruction kernel. This sensitivity negatively affects the performance of machine learning models that utilize these features. This project aims to address this challenge by investigating the effects of varying CT parameters on image-derived features and identifying optimal techniques to mitigate these effects in a task-dependent manner.

The research will pursue three interrelated innovations:

  • A novel framework for characterizing the impact of different acquisition and reconstruction parameters on QIFs and machine learning (ML) models, using patient scans with known clinical outcomes across multiple domains.
  • A systematic approach for selecting an optimal mitigation technique and evaluating the impact of normalization.
  • An open-source software toolkit, called CT-NORM, that formalizes the process of CT normalization, developed with input from academic and industry collaborators to address real-world use cases.

The project will evaluate how multiple CT parameters influence QIF values and model performance (Aim 1), assess and enhance normalization techniques (Aim 2), and engage external stakeholders to guide CT-NORM development and adoption across distinct clinical domains, including lung nodule detection, interstitial lung disease quantification, and ischemic core assessment (Aim 3). By improving the process of generating QIFs and facilitating the discovery of precise and reproducible imaging phenotypes, CT-NORM will provide the scientific community with a unified approach to characterize and mitigate variability, ultimately enhancing prediction model performance.

Contact PI: William Hsu

Funding Source: NIH NIBIB