Research

research summary

 

Positronium lifetime imaging 

Positronium lifetime carries information about the tissue microenvironment where positrons are emitted. The lifetime information can be used to understand the tissue microenvironment in vivo which could facilitate the study of disease mechanism and selection of proper treatments. However, such information has not been captured because of two technical challenges. One challenge is the low sensitivity in detecting triple coincidence events. This problem has been mitigated by the recent developments of PET scanners with long (1-2 m) axial field of view. The other challenge is the lack of proper image reconstruction methods to produce high-resolution lifetime images. We are developing new image reconstruction methods to address the second challenge. Please see our TMI paper for more details. 

Related news: Bangyan won the 1st place Best Student Paper Award at IEEE MIC for the SIMPLE method that we developed. 

NeuroExploer

The NeuroEXPLORER (NX) project is a collaboration between Yale, UC Davis, and United Imaging Healthcare America is to develop a next-generation dedicated brain PET system, funded by a $10.2M BRAIN Initiative Grant from NIH. Motivated by experience with human brain HRRT studies (Yale) and total-body uEXPLORER studies (UC Davis), the NX design focused on ultra-high sensitivity and resolution, with continuous head motion correction. The NX has a cylindrical design with a system diameter of 52.4 cm and long axial field-of-view (FOV) of 49.5 cm. The detector micro-block consists of a 4x2 array of 1.56x3.07x20 mm LYSO crystals decoded by two 3-mm SiPMs including depth-of-interaction (DOI) measurement using an optical bridge for light-sharing between axial crystal pairs. To obtain high-resolution images, we have developed a new image reconstruction method with multiple image-domain point spread functions (PSFs). 

NX team
NX team on the day of the first human scan (Jan 2023)
NX image comparison
Comparison of a human brain image obtained from the NeuroExplorer w/ and w/o MRI overlay (right panel) with that from a current state-of-the-art scanner (Vision) (left panel).

Related news: Tiantian won the 1st place PIDSC Young Investigator Award at the SNMMI annual meeting in Chicago for her multiple PSFs work.

Deep learning in PET and CT imaging

Deep learning has many applications in medical imaging from image acquisition, image reconstruction, image analysis, feature extraction, to clinical predictions. A few examples of our published work are listed here. 

  1. Deep Learning Based Joint PET Image Reconstruction and Motion Estimation
  2. Anatomically aided PET image reconstruction using deep neural networks
  3. Motion correction of respiratory-gated PET images using deep learning based image registration framework
  4. AI-Based Automated Lipomatous Tumor Segmentation in MR Images: Ensemble Solution to Heterogeneous Data
  5. Head and neck synthetic CT generated from ultra‐low‐dose cone‐beam CT following Image Gently Protocol using deep neural network
  6. Half2Half: deep neural network based CT image denoising without independent reference data

Ultra-high-temporal-resolution dynamic PET

PET is capable of monitoring the spatial-temporal distribution of injected radiotracer inside a human body in real time. Using the EXPLORER scanner, we have demonstrated dynamic PET at 100-ms temporal resolution following a bolus injection of FDG in a human volunteer. 

ultra-high-resolution-PET

See our PNAS paper:  subsecond total-body dynamic PET

Media coverages: 

  1. MedicalXpress: Look what’s inside: Full-body movies from EXPLORER scanner
  2. News Medical: EXPLORER scanner captures real-time videos of blood flow and heart function
  3. Physics World: Ultrasensitive PET enables sub-second total-body imaging

 

Parametric imaging using dynamic PET

The total-body coverage and ultra-high sensitivity of the EXPLORER scanner provide opportunities for more accurate tracer kinetic analysis in studies of physiology, biochemistry, and pharmacology. We have developed quantitative parametric image reconstruction methods for kinetic analysis and used them to analyze the first human dynamic total-body PET study. 

TB parametric images

See our JNM paper for details:  total-body dynamic PET and parametric imaging

High-Resolution Image Reconstruction

The resolution of current animal PET systems is on the order of millimeter. Images with higher resolution will greatly improve our capability of studying human diseases using mouse models. For example, sub-millimeter resolution would be required in order to detect metastases or study brain function in mice. While tremendous advance has been made in detector designs, sophisticated data processing techniques are necessary to realize its full potential. We are working on several fronts to push the resolution beyond the current limit, including super-resolution with continuous bed motion, estimation of detector blurring matrix from point source data, objective assessment of image quality for detection and quantification, and error correction in reconstruction algorithms.

We have also proposed an adaptive zoom-in PET design that incorporates a high-resolution detector into an existing PET scanner to obtain high-resolution images of a targeted region with high sensitivity and developed a method to find the optimal position of the high-resolution detector.

Shown below are some example images. Further readings: Tohme and Qi (2009), Lin and Qi (2010), Zhou and Qi (2009).
phantom zoom-in
Reconstructed images of a simulated phantom using the microPET II scanner (left) and the zoom-in PET (right). Substantial improvement in spatial resolution is achieved by incorporating a high-resolution detector into the microPET II scanner.
rat

High-resolution PET image obtained using our method (right) compared with the image obtained using the manufacture’s software (left).

Dynamic PET Data Analysis

One feature of PET is its ability to capture the dynamics of radiotracer uptake in tissue. The dynamic information has potentials to improve early detection and characterization of cancer and assessment of therapeutic response. We are developing different approaches to extract information from dynamic PET data. In (Liao and Qi 2010) we presented a segmentation method using both the spatial and temporal information to improve the classification accuracy. In (Wang and Qi 2009) we theoretically analyzed the effects of penalized maximum likelihood (PML) methods on the kinetic parameter estimation using dynamic PET. We have developed methods to reconstruct parametric images directly from sinogram data for both linear kinetic models (Wang and Qi 2009) and (Wang and Qi 2008) and nonlinear models (Wang and Qi 2009b). In addition, we also worked on methods to eliminate the needs of blood sampling in dynamic data analysis. One example was presented in (Yetik and Qi 2006) where we avoid the necessity of arterial blood sampling in dynamic PET by joint estimation of kinetic parameters and blood input function at the same time.
   
Bias images   Variance Images  
Direct reconstruction Indirect reconstruction Direct reconstruction Indirect reconstruction
bias
bias
var
var
Direct reconstruction can substantially reduce variance of parametric images without increasing bias. See (Wang and Qi 2009b) for details.