Developing a deep learning pipeline for registration of teravoxel-scale whole-brain data
BACKGROUND
Image registration is a computational method that is used in image analysis and is an essential step in the comparison of images. Registration is defined as a method of determining precise correspondences between image parameters and tissue microstructure. In neuroscience, for an integrated analysis of imaging data with atlases, precise image registration is essential, which is challenging because of stark differences in contrast across modalities. Despite having a long history of research and development of medical image registration, most state-of-the-art registration algorithms are optimization-based and do complex computation on image data iteratively until a satisfactory convergence condition is achieved. As the volume of data increases the computation time increases dramatically.
Tissue clearing (or 3D histology) is a complex biochemical process that can remove light-obstructing biological elements from tissues resulting in 3D volumes that don’t require sectioning for imaging. Volumes obtained from tissue clearing are extremely large with very high image resolution. For downscaled tissue cleared images, the best performing image registration tool ANTS takes many hours to register a pair of brain samples.
Deep learning (DL) has become popular in medical imaging and has achieved very promising performance in tasks such as object detection and organ segmentation. However, the direct application of deep learning-based models in image registration is not straightforward due to the complex nature of image registration and the lack of ground-truth registration data. In deep learning-based image registration methods, the algorithm learns a parametrized registration function from a collection of volumes. The aim of this project is implementing a deep convolutional neural network (CNN), that takes two n-D input volumes and outputs a mapping of all voxels of one volume to another volume.
PROJECT DESCRIPTION AND DETAILS
The student will be involved in developing and evaluating a novel DL based pipeline for registration of brain tissue cleared data to address the above-mentioned challenges. To train and evaluate this pipeline, a valuable dataset is necessary. We have access to a very large (whole-brain) mouse and rat 3D microscopy data (tissue clearing). This pipeline enables for an integrated analysis of cleared volumes and imaging data with atlases and will be used in combination with other pipelines that have been developed by our laboratory for different image processing tasks
such as a novel computational pipeline for segmentation neurons based on convolutional neural networks. The student will be supervised by Dr. Maged Goubran and Dr. Bojana Stefanovic.
STUDENT BENEFITS
- Exposure to 3D multi-label whole-brain microscopy dataset
- Hands-on experience with implementation of CNN and DL
- Hands-on experience with analyzing, evaluating, and optimizing the state-of-the-art methods in DL
- Access to computational resources needed to accomplish the aims
- Opportunity to write a conference abstract and involvement in resulting papers
- Learning cutting-edge data-driven techniques in image processing
- Journal club discussions and scientific paper reading
- Engagement with other students and projects in the lab
OUR LABORATORY
Dr. Goubran’s lab develops novel computational, machine learning & imaging tools to probe, predict and understand neuronal and vascular circuit alterations in neurological disorders, including Alzheimer’s disease, stroke, and traumatic brain injury. Dr. Stefanovic’s lab is focused on developing tools for studying the inner workings of the coolest organ of the body and examining new ways of supporting recovery processes in Alzheimer’s disease, stroke, and traumatic brain injury. Both labs are located at the Sunnybrook Research Institute and consist of a multidisciplinary team of engineers, neuroscientists, and software developers, who are responsible for the development and implementation of tools to quantify various imaging markers of brain disease.
ELIGIBILITY
- Proficiency in programming (specifically python) ideally familiar with libraries such as TensorFlow, PyTorch and Keras.
- Able to work 8-10 hours/week with potential for extension.
- Familiarity with bash and Compute Canada/SciNet
- Advantage: One of CSC303, CSC311, CSC320, CSC336, MAT332, and MAT344 or CSC494/495
CONTACT INFORMATION
For further information and to apply please contact our graduate student Ahmadreza Attarpour a.attarpour@mail.utoronto.ca