Assistant Professor, Division of Medical Physics, Department of Oncology, University of Alberta
Senior Medical Physicist, Dept. of Medical Physics, CCI
PhD Medical Physics: University of Alberta, 2013
Member: Canadian College of Physicists in Medicine
Fellow: Canadian College of Physicists in Medicine
The objective of my research is to enhance the geometric accuracy of radiation therapy (RT) for cancer patients, especially for treating mobile tumours (lung, liver, or prostate tumours) using a hybrid radiotherapy-MRI system called Linac-MR. To achieve this, I have established a novel RT method called nifteRT (non-invasive intra-fractional tumour-tracked RT) since 2007 producing more than 20 refereed journal publications. NifteRT requires 4 consecutive steps continuously run during irradiation.
Step 1: Fast intra-fractional MR imaging Fast intra-fractional MR imaging (i.e. imaging during irradiation) is the first step of nifteRT to localize a moving tumour. Various incoherent partial k-space acquisition schemes were explored to reduce the scan time, combined with novel image reconstruction algorithms: prior data assisted compressed sensing (PDACS), sliding window PDACS, CS and principal component analysis (CS-PCA). Recently, I developed a new algorithm using fully convolutional neural networks (FCNN), which is a type of deep-learning neural network (DLNN) that is applicable to MR image reconstruction tasks.
Step 2: Tumour auto-contouring Accurate tumour auto-contouring from the intra-fractional MR images is the most critical component of nifeRT, because it dictates the accuracy of tumour localization. I have developed several algorithms to detect the shape and position of the tumour from continuous MR images. Several different approaches were studied for tumour auto-contouring including histogram shifting, pulse-coupled neural networks (PCNN, a 2D neural network that is applicable to image segmentation), and U-net (a type of DLNN applicable to image segmentation).
Step 3: Tumour motion prediction Tumour motion prediction occurs immediately after the auto-contouring that localizes the current tumour position. The prediction is necessary to compensate for the tumour motion during the inevitable time delay; the time interval between the detection of current tumour position and the radiation beam adjustment upon the multi-leaf collimator (MLC) reaching its desired leaf position. I adopted a long short-term memory (LSTM) in nifteRT, which is a type of DLNN applicable to time series prediction.
Step 4: Real-time MLC control Immediately after the motion prediction, the MLC is driven to conform the radiation beam to the auto-contoured tumour shape at its predicted position. I built the software/hardware components of the real-time MLC control system. Continuous operation of MLC is required during nifteRT, in order to adjust the radiation beam on-the-fly to track the moving tumour.
I conducted the world's first demonstration of nifteRT in 2013, utilizing the 2D intra-fractional MR imaging capability of our prototype linac-MR. I integrated all software/hardware components developed for Steps 1 to 4 into a single nifteRT system. While my motion phantom was reproducing 1D lung tumour motion within our linac-MR, my nifteRT system was able to (1) acquire intra-fractional MR images of a moving target at 4 fps, (2) automatically detect the shape and location of the moving target from each MR image, (3) predict the future target position over the time delay, and (4) control the MLC on-the-fly to conform the radiation beam to the auto-contoured target shape at its predicted position.
One critical component of nifteRT is the ability to accurately calculate the radiation dose absorbed by the moving tumour and surrounding organs. This will be done using Monte Carlo (MC) simulation, a widely accepted dose calculation technique in RT. Whereas typical MC simulations are done in static 3D volumes, my research centers on a moving target which necessitates the use of 4D (3D space plus time component) simulations. These simulations will run on the TOPAS (Tool for Particle Simulation) application, which is specialized in simulating moving objects during radiation delivery.
The University of Alberta, its buildings, labs, and research stations are primarily located on the traditional territory of Cree, Blackfoot, Métis, Nakota Sioux, Iroquois, Dene, and Ojibway/Saulteaux/Anishinaabe nations; lands that are now known as part of Treaties 6, 7, and 8 and homeland of the Métis. The University of Alberta respects the sovereignty, lands, histories, languages, knowledge systems, and cultures of First Nations, Métis and Inuit nations.
