Medical imaging with the a focus on:

• Image segmentation
• Automatic lesion detection
• PET lesion synthesis
• Image perception
• Quantification of physiologic function using dynamic imaging
• Motion detection and correction
• Quantitative SPECT
• Prediction of kidney failure
• Prediction of Tacrolimus dosing

Current projects:

Myocardial blood flow quantification with ⁸²Rb PET

At the University of Ottawa Heart Institute, National Cardiac PET Centre we have been developing PET for routine quantification of myocardial blood flow (MBF) and myocardial flow reserve (MFR). These efforts have resulted in the development and commercialization of the following technologies:

• ⁸²Rb/⁸²Sr generators and infusion system technologies (licensed to Jubilant-DraxImage and commercialized as Ruby-Fill^®)
• FlowQuant^TM image analysis software of which some of the MBF and MFR technologies have been been licensed to INVIA Medical Imaging Solutions.

Our current efforts are focused on improving precision of MBF and MFR by tackling problems such as respiratory and patient motion correction. Collaborators:

• Robert. deKemp, PhD, University of Ottawa Heart Institute, Cardiac PET Centre
• T. Xu, PhD, Carleton University, Department of Physics
• A Alessio, PhD, Michigan State University, Department of Computational Mathematics, Science, and Engineering; Biomedical Engineering and Radiology; Institute for Quantitative Health Science & Engineering

Students:

• Spencer Manwell, PhD, Department of Physics, Carleton University, Thesis
• Chad Hunter, Postdoctoral Fellow
• Anbhu Sritharan, MASc, Department of Systems and Computer Engineering, Carleton University, Thesis
• Simisani Takobana, MASc, Department of Systems and Computer Engineering, Carleton University, Thesis

Lesion synthesis in FDG PET and CT

We have developed state-of-the-art methods for synthesizing realistic lesions in both PET and CT data. We are currently applying these methods to generate large libraries of images with well characterized fake lesions for the following goals:

• Characterize the limits of detection (LOD), including lesion contrast, lesion size, image noise and anatomical region, of human-observers under fixed imaging conditions such as PET image reconstruction method using synthetic lesions.
• Optimization of PET image reconstruction and visualization for lesion detection on the merits of LOD.
• Developing machine-learning based lesion detection artificial intelligence (AI) using very large libraries of synthetic lesions.
• Objectively comparing the performance of clinicians and AI for the task of lesion detection with regards to both success rates and LOD.

Collaborators: Charles Collin, Elizabeth Krupinski, General Electric Health Care PET R&D team,

Student:

• Hanif Gabrani-Juma, MASc, Biomedical Engineering, Carleton University, Thesis
• Quinn de Bourbon, MSc, Medical Physics, Carleton University, Thesis

Enhancement, Segmentation, and Classification of Lung Ventilation and Perfusion Scintigraphy

In collaboration with Dr Eric Moulton of Jubilant-Radiopharma, we are developing AI for interpretation of lung V/Q studies.
Students:

• Siraj Ghassel, MASc, Computer Science, University of Ottawa, Thesis
• Amir Jabbarpour, PhD Candidate, Medical Physics, Carleton University

Prediction of Kidney Failure and Optimal Therapy

These collaborative efforts include amongst others nephrologists Ayub Akbari and Gregory Hundemer, Biochemist Chris McCudden and Engineering Professor James Green. This group is developing predictive algorithms for timely selection of patients with chronic kidney disease in the Multicare Kidney Clinic that are likely to require renal replacement therapy in the near-future. Furthermore, we are endeavoring to develop algorithms to guide clinicians for optimal Tacrolimus dosing in renal transplant recipients.

Students: 

• Martin Klamrowski, MASC, Electrical and Computer Engineering, Carleton University, Thesis; and PhD Candidate
• Elmira Amooei, PhD Candidate, Biomedical Engineering, Carleton University.

Recent Projects

FDG PET-CT Image segmentation and lesion detection

Using analytical methods and machine-learning we are developing artificial intelligence (AI) for computer aided diagnosis (CAD) of FDG PET-CT studies. In collaboration with Hermes Medical Solutions we are enabling collaboration between human and machine observers using Hybrid3D^TM to improve diagnostic accuracy, automatically generate templated clinical reports, and to provide active learning feedback to our AI.

Past Student: Odai Salman, PhD, Systems and Computer Engineering, Carleton University, Thesis.

 

Two-compartment exchange phantom

With Catherine Coolens and Shelley Medical Solutions we have developed and validated a two-compartment exchange phantom for multimodality imaging. The DCE perfusion from phantom phantom can be used for validation of quantitative dynamic imaging including PET, SPECT, CT and MRI for modality development, qualification testing in clinical trials, and quality control.

 

Collaborators: C. Coolens
Student: Hanif Gabrani-Juma, BSc, Department of Systems and Computer Engineering, Carleton University
Industry partner: Shelley Medical Imaging Technologies

 

Quantitative SUV SPECT

Quantitative SPECT that can generate SUV scaled images has become routinely available, but its clinical utility is not yet clear. Since 2015 we have been increasingly applying SUV SPECT imaging in routine clinical practice to enable research of its clinical utility. To date we have produced numerous publication and abstracts using SUV SPECT and parathyroid adenomas, thyroid imaging and tumor imaging.

Collaborators: L. Zuckier, W. Zeng Industry partner: Hermes Medical Solutions

 

Matlab Source Code

Image Perception and Lesion Synthesis

Two separate, but related, tools used in our image perception and limits-of-detection (LOD) work in PET/CT.

• The Lesion Synthesis Toolbox is an integrated workflow for generating well characterized and realistic lesions in PET and CT data. While the tool was designed with the intention of being vendor neutral, for the time being it was only implemented and tested on General Electric Healthcare products and relies on the PET Duetto image reconstruction toolbox available from GE through a research collaboration agreement. Functionality includes a graphical user interface (GUI) for data retrieval from the scanner console, PET image reconstruction, definition and batch generation of user defined lesions and building of an image database of synthetic lesion and their ground truth.
• The image perception suite enables a researcher to conduct and analyze an image perception study. A library of images and ground truth data (e.g. synthetic lesions) is used to run the study, collect user responses and then perform initial analysis such as deriving performance metrics (e.g. LOD).

The work was first described in Juma and Klein 2020. Both tools may be run as an application on a personal computer or as a web-hosted service. The latter enables to host the service on a single server computer that is accessible on the network, reducing setup and maintenance time and leveraging computing power to process data in the cloud.
We've since demonstrated the use of these tools for characterizing the performance of lesion detection AI in de Bourbon and Klein, 2025.

Bland-Altman and Correlation Analysis

Open source tool for plotting Bland-Altman and Correlation plots and performing statistical analysis.

View4D

An interactive image viewer for 3D and 4D data. It is intended for visualization and interaction with tomographic medical images. The 4th dimension can represent time, phase, energy or any other image component. Time-activity curves of the user selected pixel are displayed and the plot can be used specify the time range to summarize and display. Fused image overlay with a second volume (3D) image and mesh intersection contours are optional features.
Input argument can be used to customize/initialize the display. Callback function can used to integrate the viewer into custom applications. The viewer can be called in wait-for-close or continue-execution mode.
Demo call functions and data are included in the package along with instructions.

VolumeViewer

Similar to View4D in many ways, but uses a single display of intersecting slices through a volume. Produces prettier images, but is typically less functional than View4D. This interactive image viewer is intended to explore multi-frame tomographic medical image data consisting of 3 spatial domains and a 4th domain (e.g. time, phase, energy). Input argument can be used to initialize the display. A callback function can be specified to trigger on pixel selection and it can be run in wait-for-close mode, in which the user can select a cropped volume, 2D slices, and a range of time frames (4th dimension). So it is simple to integrate into other programs.

DICOM Convert Server

Reading DICOM image series into Matlab can be challenging due to a lack of standardized format among vendors and can be slow. The DICOM Convert Server is a free executable that monitors an incoming directory and converts DICOM series into a single matlab (.mat) file containing the image volume and header information. Detailed instruction are included in the accompanying manual.

Other Contributions

Glomerular Filtration Rate Electronic Worksheet Template

The GFR spreadsheet incorporates quality control indicators as described in NMC 2019:40(1):30-40. It is designed to be used in a routine clinical setting to calculate GFR and generate a printed report as part of a complete patient workup. The spreadsheet can be tailored to site's needs. Important note: By using our tool, you agree to these terms and conditions, made between you and the Ottawa Hospital Research Institute and the Ottawa Hospital. The template should be tailored to a clinic's needs and is provided solely as a guidance template. The template is provided as is, with no guarantees or warranties, and its use is at your own risk. It is not a licensed medical device in any jurisdiction.
The Ottawa Hospital Research Institute, the Ottawa Hospital, and its researchers, including Ran Klein, expressly disclaim all warranties of any kind regarding this tool, either express or implied, including, without limitation, warranties of title, non-infringement, and implied warranties of merchantability or fitness for a particular purpose. The Ottawa Hospital Research Institute, the Ottawa Hospital, and its researchers, including Ran Klein, will have no liability for any damages, injury or any other loss whatsoever suffered as a result of use of, or reliance upon, this tool. In the case of loss, injury or damage of any kind to a person resulting from your use of this tool, you agree to indemnify and hold harmless the Ottawa Hospital Research Institute, the Ottawa Hospital, and its researchers, including Ran Klein, from any and all claims arising therefrom.

Commercial Medical Products

RUBY-FILL® (Rubidium Rb82 Generator and Infuser)

Developed as during my Master's thesis work, the infuser delivers a controlled activity profile to the patient which is optimized for dynamic cardiac imaging on 3D PET scanners, to avoid saturation of the detectors at early phases of the scan and ensuring ample activity for late phases of the scan. Hence both high quality myocardial perfusion images generated and precise myocardial blood flow measurements can be made from a single acquisition, increasing clinical throughput, reducing radiation exposure, and providing encompassing diagnostic information. The system also offers unique quality and safety features that enforce rigorous radiation protection while extending the useful life of the ⁸²Rb/⁸²Sr generator.

4DM Myocardial Blood Flow Quantification Software

Dynamic image analysis algorithms developed and validated at the during my PhD and Research Associate day at the University of Ottawa Heart Institute, National Cardiac PET Centre, have been implemented and commercialized by INVIA Medical Imaging Solutions. The Coronary Flow Reserve (CFR) package processed dynamic PET and SPECT images to measure regional myocardial blood flow and coronary flow reserve to improve diagnosis of ischemic disease that may be missed by conventional imaging technology.