Highly accurate photon-counting CT for cancer imaging with deep-learning image reconstruction

NAISS 2023/22-85


NAISS Small Compute

Principal Investigator:

Mats Persson


Kungliga Tekniska högskolan

Start Date:


End Date:


Primary Classification:

20603: Medical Image Processing




Cancer is one of the leading causes of mortality with an expected 10 million deaths annually. A very widely used imaging modality for diagnosing cancer is x-ray computed tomography (CT), which provides three-dimensional images of the human body is reconstructed from x-ray measurements. An emerging technology within cancer imaging is radiomics, where quantitative numerical metrics are derived from measurements on cancer tumours and used to train machine-learning algorithms to predict the disease trajectory of the patient. Despite its high usefulness, there are limitations with the current CT technology with respect to diagnostic quality and quantitative accuracy. The emerging photon-counting CT technology can overcome these limitations with its higher spatial resolution, lower image noise, and improved material-selective imaging. This is particularly true for imaging cancer, since developments in radiomics are hampered by the imperfect quantitative accuracy of today’s CT technology. However, new image reconstruction methods need to be developed in order to achieve the full potential of the technology. Deep-learning-based image reconstruction, a new technology for image CT reconstruction, has demonstrated substantial image quality improvement and fast reconstruction. We will develop a deep-learning-based CT image reconstruction method that is especially suited for generating highly accurate photon counting CT images together with maps of image uncertainty. In order to do this, we will use NAISS compute resources to train deep neural networks that are able to map measured x-ray imaging data into images with as high accuracy and resolution as possible. The image data that will be stored on NAISS compute resources will consist of images of test objects (“phantoms”) and of anonymized datasets retrieved from internet databases, i.e. no protected health information will be stored on NAISS servers. After training, however, we will download the trained models to our in-house local computers and apply them to patient images acquired with a photon-counting CT scanner prototype developed in our lab. As part of the evaluation process, we will investigate the usefulness of the new imaging technique for diagnosis and radiomic characterization of tumours. We will generate 3D-printed phantoms resembling human anatomy with known shape and composition, scan these with a photon-counting CT scanner, use the novel image reconstruction method to generate images and extract radiomic features from these images. This will allow us to investigate how the accuracy of the radiomic features is related to the characteristics of the deep-learning-based reconstruction algorithm. This project will complement a related project called “Deep-learning data processing for photon-counting CT” that our lab is running on the Berzelius cluster, but with a different focus: developing a quantitatively accurate reconstruction algorithm for cancer radiomics. The anticipated outcome is that photon-counting spectral CT with deep-learning reconstruction can give drastically improved diagnostic quality and radiomic measurement accuracy without extra dose. This can lead to saved lives and new research avenues in the field of data-driven cancer diagnosis.