Enhancing PET-CT workflow and data quality

Improved PET/CT clinical workflow and productivity with AI-enabled reconstruction and processing methods

Area of work


Imaging modality


Size of data set


Dr Fergus Gleeson, NCIMI CMO.

Project lead

Professor Fergus Gleeson

Industry partners

Number of NHS partners


Unmet need

The radiology workforce is overstretched and understaffed, with the latest Royal College of Radiologist Census indicating existing shortfalls in the workforce of as much as 44%. Supporting enhanced workflow and increased data quality from scans can help address existing work backlogs and support radiologists in focussing on the most critical cases.

Machine learning can be used to improve the efficiency of the diagnostic process, improve the quality of data gained from the process and find deeper patterns within the data. All of these will improve the value to the clinician and patient, enabling earlier and more accurate decisions as well as better overall patient outcomes. It should be noted that emphasis will be placed on solutions that improve performance, time, workflow or cost – or ideally all of these. 

Project aim

The study aims to develop and implement machine learning enhanced methods that improve workflow and data quality for PET/CT imaging studies and to produce improvements in healthcare delivery that reduce time and patient radiation dose and increase quality and accuracy of diagnoses while improving the overall workflow.

Machine and deep learning (ML, DL) have shown the potential to improve both efficiency, accuracy and quality of health care delivery and workflow. There are many potential areas where ML or DL can interact, from pre-population of information for a patient scan based on prior information, improvement of the quality of data produced by an imaging system, or deeper pattern recognition applied to the data from a procedure.

All of these areas are currently impacted by the complexity of today’s modern systems – and while the capability of current systems has never been higher, it often comes at the cost of more burden on those who use the systems.

A simple example is the production of more and better imaging data, often read in combination with data from other modalities and prior scans to form a current action for the patient.


  1. Reduce dose and/or scan time for PET/CT by utilizing ML-based data enhancement. Such enhancement may include de-noising of lower dose or shorter scan time PET data. 
  2. Enable multi-gate respiratory motion correction for PET/CT for whole-body imaging. Automated PET respiratory motion correction can improve image quality and quantification. Machine learning will be used to improve the motion estimation by reducing noise in the time-constrained gated image data and therefore increase the accuracy of a 3D, non-rigid registration as compared to the current approach.
  3. Decrease PET image reconstruction time for the state-of-the-art Q.Clear method by using ML to reduce the number of required reconstruction iterations. This is an important factor by itself but becomes even more important in combination with whole-body automated all-counts respiratory motion correction or whole-body dynamic PET imaging. 
  4. Decrease scan prescription, setup and acquisition time and improve the patient experience by utilizing ML techniques to eliminate or optimize the tasks performed during a typical PET/CT Scan. 

No articles found