The Biomedical Signal and Image Computing Lab (BiSICL) is a multidisciplinary research facility at UBC dedicated to computational research in biomedical imaging. Our lab's main objectives are to create, develop and translate innovative techniques for automated processing, analysis, understanding, and visualization of structural and functional medical imaging data so that they can be applied in a clinically-focused, disease-specific manner. The uniqueness of the research conducted at BiSICL stems from the close and direct collaboration with local and international clinicians of the highest caliber and on-site presence in leading hospitals and clinical sites spanning literally from the lab to the bedside.
Applications of our work range from basic imaging biomarker based study of disease and recovery mechanisms to emerging image-guided surgical interventions where innovations in computational methods are essential for enabling, advancing, accelerating and enhancing the quality and efficacy of health research and clinical practice. Currently our projects focus on neuroimaging (functional and diffusion MRI ), robotic surgery (abdominal cancer) and novel uses of 3D ultrasound (orthopedics). Our research promises substantial benefits to large sections of the world's population especially with our aging demographics where related diseases are common, debilitating and pose a tremendous burden and cost on healthcare systems.
A list of our publications is available here.
Our objective is to overlay data derived from pre-operative CT images onto the surgeon’s stereo endoscopic view. Moreover, the pre-operative data will be first positioned and then deformed (e.g. stretched, bent, or cut) continuously to match the current state of the patient’s anatomy. Lastly, left and right (stereo) projective views of the overlaid data will be streamed onto the left and right camera views provided to the surgeon console, thus augmenting the 3D endoscopic view of the operation field. This MIS enhancement will radically improve the surgeon’s experience and efficiency, increase the precision of the surgery, decrease the surgery time, and reduce collateral damage, which in turn will lead to improved surgery outcomes and patient recovery.
Although it is now well established that brain fiber pathways serve as the physical substrate for functional interactions, this information is rarely exploited in current fMRI studies. In this project, we investigate the implications of fusing information regarding brain structure and brain activity in the presence and absence of explicit input. This multimodal approach enables us to analyze how the brain responds to stimuli around its baseline and to regularize the analyses of functional brain dynamics using structural connectivity information. The methods we develop ultimately serve to enhance our general understanding of the human brain organization by shedding light on the structure-function relationship in the brain.
In this project, we are investigating extracting the 3D biomechanical patient-specific model of the tongue tissue mostly from magnetic resonance images with specific focus on simulating the swallowing process. The successful solution will enclose an efficient and reasonably fast combination of different image processing techniques with minimum user interaction, aiming to be included in a computerized visual platform like Artisynth, in order to help in visualizing the patient-specific anatomy and dynamics of the tongue as well as predicting the possible effects of any treatment or surgery in the overall functionality of the oropharyngeal structures in case of any swallowing disorders.
Orthopedic imaging has traditionally relied on ionizing radiation based modalities such as x-ray, fluoroscopy and CT. We are investigating the employment of 3D US as an alternative safer imaging modality for a prospective minimally invasive computer assisted surgery system designed specifically for pre-operative bone fracture assessment and intra-operative guidance in fracture reduction procedures. We have demonstrated fracture detection and have developed US-CT fusion methods to improve intra-operative visualization of bone surfaces.
fMRI has become the dominant imaging modality for studying human brain function non-invasively. Most fMRI studies have focused on inferring changes in brain activity from fMRI signal intensity modulations. Here at BiSICL, we are extending traditional fMRI analysis methods by designing new robust spatial descriptors to characterize brain activation pattern in areas implicated in neurodegenerative diseases such as PD. In particular, we are exploring the use of invariant spatial features to examine the spatiotemporal properties of activation within various brain regions of interest (ROIs).
In addition to inferring brain activation, fMRI is often used to study the functional integration of different brain regions. The apparent inter-subject variability, however, renders detection of representative group networks very challenging. Therefore, we are currently designing novel approaches based on sparse multivariate models that integrate group information in detecting common brain networks across subjects. We are also investigating the implications of inferring functional connectivity based on spatial modulations of blood oxygen level dependent (BOLD) signals in contrast to traditional mean region of interest (ROI) intensity time courses.
Standard fMRI analysis approaches typically examine each voxel in isolation despite that each voxel is unlikely to function independently. In particular, numerous past studies have shown the presence of task-related functional networks, which suggests that functional connectivity may be another indication of brain activity. Here at BiSICL, we are designing novel methods to incorporate this connectivity prior into the activation detection procedures. Specifically, we are employing sparse multivariate models and graphical models that enable connectivity information to be seamlessly integrated into the activation statistics.
Regularization plays a crucial role in improving the robustness and applicability of image segmentation techniques. Through the use of weighted regularization terms in conjunction with data fidelity terms, images plagued by high levels of deterioration, i.e. noise or poor edge contrast, are prevented from causing excessive irregularities and inaccuracies in the resultant segmentation. Conventionally, the associated parameters are set through tedious trial and error procedures and kept constant over the image. However, spatially varying structural characteristics combined with varying noise and imaging artifacts significantly complicate the selection process of segmentation parameters. At BiSICL, we are developing novel approaches for automating parameter selection by employing data cues to prevent excessive regularization of key features and important image regions.
State of the art medical imaging technologies provide numerous high quality structural and functional data invaluable for medical practice and research. MRI and fMRI are commonly used examples. However, such data is typically enormous in size and pose significant demands on disk storage and channel bandwidth requirements. At BiSICL, we are developing highly efficient lossless 3D and 4D image compression techniques based on symmetry and motion compensation concepts to remove data redundancy.
Analysis of vessel-like anatomical structures such as vasculature is important in many medical applications including tumors, aneurysms, diabetes, and hypertension. However, vessel segmentation is especially difficult due to the complexity of associated topologies and poor image quality especially that of thin vessels. We are developing novel approaches for such segmentation tasks that allows the user to incorporate various degrees of interactivity vs. automation in 2D and 3D.
Most anatomical structures and regions of interst captured with various 3D medical imaging technologies are of complex shape and topology. Due to this as well as the commonly encourntered poor image quality and pathology, accurate yet robust segmentation of such structures is generally extremely difficult to achieve using fully automatic methodologies. On the other end of the spectrum, manual segmentation alternatives are very tedious and impractically time-consuming. At BiSICL, we are developing highly-automated 3D segmentation tools that incorporate human knowledge through minimal user interaction that guides 3D segmentation tasks in a very intuitive and efficient manner improving accuracy and robustness.
Voxels are typically a discrete representation of a continuous scene and as such are not independent; similar regions exhibit similar properties such as intensity or texture and therefore can be grouped so that the complexity and computational cost are reduced from the number of voxels to the number of regions observed in the image. At BiSICL, we are developing new region-based models in a hidden Markov model (HMM) framework and novel tree-structured parameter estimation algorithms for efficient and accurate 3D image segmentation.
Orthopedic imaging has traditionally relied on ionizing radiation based modalities such as x-ray, fluoroscopy and CT. At BiSICL, we are investigating the use of 3D US as an alternative safer imaging modality for a prospective minimally invasive computer assisted surgery system designed, specifically for pre-operative bone fracture assessment and inter-operative guidance in fracture reduction procedures.
Bone diseases such as arthritis and osteoporosis are very painful diseases that severely limit movement afflicting many, especially the elderly. At BiSICL, we are working on novel techniques that help determine the nature and severity of such diseases, as well as help devise appropriate treatments in a non-invasive manner. To achieve that, we employ different medical imaging modalities including CT, MRI, and fluoroscopy and design schemes for data fusion that enable the integration of the different but complementary information extracted from each modality.
Research in the area of spinal cord injury has traditionally been carried out through biomechanical testing where animal spinal cords are exposed and subjected to mechanical injury. However, a major limitation of this is the need to surgically expose the cord, which changes the physiological environment and mechanical boundary conditions of the spinal cord. We are developing alternative approaches based on MRI imaging for the non-invasive study and quantitative assessment of spinal cord, in vivo, without exposure. Our methods allows for the study of cord deformations such as myelopathy-related sustained compression in its natural physiological environment. This work is done in collaboration with the Injury Biomechanics Laboratory at UBC.
MRI affords neurologists an invaluable non-invasive means for examining patients with neurodegenerative diseases such as MS, where brain atrophy has been identified as one of the important biomarkers tracking disease progression. We are developing new approaches to fully automate the brain segmentation process on both normal and degenerated brains based on hybrid probabilistic and geometric segmentation paradigms. This work is done in collaboration with the MS/MRI Research Group at UBC.
Analysis of vasculature is important in many medical applications including tumors, aneurysms, diabetes, and hypertension. However, vessel segmentation is especially difficult due to complexity of associated structures and frequent poor image quality especially that of thin vessels. We are developing an interactive approach to segmenting vasculature that is globally optimal and most importantly, that ensure topological fidelity. This work is done in collaboration with the Medical Image Analysis Lab (MIAL) at SFU.
The completion of the Human Genome Project marked an important initial step in the exploration of how genetic variations interact with the environment to confer individual resistance or susceptibility to disease and responsiveness to medical interventions. We are working on new microarray-based genotyping strategies capable of providing rapid genotyping for prospective clinical trials and clinical practice, with a turn-around timeline compatible with clinical decision-making. This involves designing an automatic microarray analysis system for data gridding, segmentation, and gene calling that is operating in real time. This work is done in collaboration with the iCAPTURE Centre at St. Paul's Hospital.
In fMRI, movement of the head during the imaging process can lead to erroneous data that do not belong to the specific activation area corresponding to a task. It is crucial to correct for this problem in order to guarantee correct further data analysis. In the adjacent top figures, the coloured areas indicate surface areas that are labeled as active. This is due to motion distortion since, in this case, there was not supposed to be any activation at the brain edge. In order to overcome this problem, registration and motion correction techniques are used, leading to results as shown in the bottom figures. We are working on a motion correction technique such as MCICA by using multiple reference images to improve accuracy and also to account for non-rigid distortion, and also on Maximum Entropy based techniques to achieve the same goals.