Past talks

Date: Monday, December 19th, 2011, 3-4pm

Location: Princess Margaret Hospital, Level 1B, Red Room, 610 University Avenue, Toronto, Ontario (map)

Moderators: Doron Dekel, Claron Technology and Vladimir Pekar, Philips

Title: Editing 3D Segmentation Results - Discussion

Description: Often, the results of automatic segmentation results do not fully satisfy the user. Editing the segmentation contours slice-by-slice is very time consuming and negates much of the benefits of the initial automation. The purpose of this meeting is to share ideas regarding how to make such editing as efficient as possible. To kick off the discussion, Claron and Philips will present and demonstrate some clinical applications and editing techniques they have developed to date.

Date: Monday, November 7th, 2011, 2-3pm

Location: Princess Margaret Hospital, Level 1B, Red Room, 610 University Avenue, Toronto, Ontario (map)

Presenter: Mark Taylor, University Health Network

Title: Commercialization of Med-Tech Research at Hospitals

Description:

Date: Monday, October 31st, 2011, 2-3pm

Location: Princess Margaret Hospital, Auditorium 6-604, 610 University Avenue, Toronto, Ontario (map)

Presenter: Anton Semechko, University of Guelph

Title: Development of a Multi-body Statistical Shape Model of the Wrist Joint

Description: With continually growing availability of high performance computing resources, the finite element (FE) methods are becoming increasingly more efficient and practical research tools. One of the research interests of our group is concerned with the development of a detailed, anatomically accurate, FE model of the human hand and wrist. As a first step in this direction, we have employed a population-based modeling strategy to create a composite, statistical shape model (SSM) of the carpus. In addition to circumventing the main shortcoming of subject-specific biomechanical models (of having limited generalization ability), our model possesses a wider range of biomedical applications that include the optimal design of prosthetic and fixation devices; the clinical reference of normal carpal variations and inter-carpal relationships, and the systematic study of structural changes in bone anatomy at various stages of growth or disease. In this presentation, I will talk about the techniques developed during my Masters, which were used in construction of the composite SSM of the carpus. The methods, however, are directly applicable to other multi-body joints, as well as single-body anatomical structures topologically equivalent to a sphere (e.g. brain, lungs, kidneys, liver).

Date: Monday, October 17th, 2011, 3-4pm

Location: Princess Margaret Hospital, Auditorium 6-604, 610 University Avenue, Toronto, Ontario (map)

Presenter: Dr. Hamid Tizhoosh, Segasist Technologies

Title: Reconcillio - Overcoming Variability in Contouring

Description: In contouring medical images, there is generally no “gold standard”, meaning that there is no 100%- accurate contour for a given lesion/organ of a given patient. Different experts (radiologists, oncologists, pathologists etc.) contour differently. This well-known dilemma is called “inter-observer variability” and constitutes a serious impediment in medical imaging. The same problem also applies to individual experts, who can mark the same image differently when they observe it for a second time. This is called "intra-observer variability". Segasist Reconcillio is a revolutionary approach to the auto-contouring of medical images. It uses the Segasist Engine to create “Knowledge Maps” for each expert user who works with the software. Over time, the Knowledge Maps grow and converge toward a high agreement with the expert’s expectations, in terms of where the contour should be. Hence, multiple contours can be extracted, reflecting the individual differences; a consensus contour can be built and offered to each individual user to assess the quality (accuracy) of his/her marking. Reconcillio can also assist the same clinical expert to become more consistent in the way he/she contours. The verification by Reconcillio can offer each expert a consistent contour by employing the Knowledge Maps of that user.

Date: Monday, August 22nd, 2011, 2-3pm

Location: Princess Margaret Hospital, Room 7-605,610 University Avenue, Toronto, Ontario (map)Date:

Title: Medical Image Processing In Action

Description: Live demos and presentations of medical image processing systems, followed by networking

Demos/Presentations:

  • Title: Rapid deformation image registration on the GPU for adaptive radiation therapy

  • Description: A high performance deformable image registration algorithm on the standard graphics process unit (GPU) was developed to compensate for the inter-fraction and intra-fraction deformation in fractionated radiotherapy. The deformable registration algorithm was formulated under a variational framework and numerically solved using a block-matching algorithm. The ultimate goal is to enable adaptive dose computations at the treatment console.

  • Kevin Wang

  • Research Associate

  • Radiation Medicine Program

  • Princess Margaret Hospital

  • University Health Network

    • Title: VURTIGO: Visualization Platform for Real-time, MRI-guided interventions

    • Description: Vurtigo is a four-dimensional (3D + time) real-time visualization software for guiding cardiovascular interventions. It is designed to be part of a pipeline that can connect it to a magnetic resonance imaging (MRI) scanner, actively tracked catheters, and navigational devices. In particular, we will be demonstrating an example of a cardiac electroanatomic mapping session with an actively-tracked catheter under MR.

    • Roey Flor, Labonny Biswas, Samuel Oduneye, Venkat Ramanan, Kevan Anderson, Stefan Pintilie, Perry Radau, Graham Wright

    • Wright Group Imaging Research

    • Sunnybrook Research Institute

    • Title: B-CAD Demo

    • Description: B-CAD is the first breast ultrasound computer-aided detection (CAD) tool to provide advanced decision-making support for lesion analysis. Using B-CAD, radiologists are able to consistently document notes, patient history, and lesion information, resulting in a thorough diagnosis and efficient communication.

    • Frederic Lachmann, PhD

    • CTO The Medipattern Corporation

    • Title: The UHN X-Eyes Software Platform

    • Description: The demo will present the custom 3D visualization and navigation platform developed by the UHN Guided Therapeutics (GTx) program for clinical applications in image-guided surgery, radiation therapy, and interventional radiology. Built on open-source toolkits, the software provides registration and visualization functionality integrating 3D imaging (MR/CT/CBCT), planning contours, real-time tracking (IR/EM), and endoscopic video.

    • Michael Daly, Institute of Medical Science, U of T

    • Jimmy Qiu, Ontario Cancer Institute, UHN

    • Harley Chan, Ontario Cancer Institute, UHN

    • Collaborators: Robert Weersink, Radiation Physics, UHN Walter Kucharczyk, Medical Imaging, UHN Jonathan Irish, Surgical Oncology, UHN David Jaffray, Radiation Physics, UHN

    • Date: Monday, July 11th, 2011, 2-3pm

    • Location: Princess Margaret Hospital, Room 7-605,610 University Avenue, Toronto, Ontario (map)

    • Speaker: Olesya Peshko, McMaster University

    • Title:

    • Automatic Localization of Fiducial Markers in X-ray Fluoroscopy Images for Tracking of Organ Motion

    • Abstract:

    • Image-guided radiation therapy with its ability to visualize soft-tissue structures of cancer patients has revolutionized the delivery of external beam radiation treatments. Still, one of the major challenges in this field is intra-fraction organ motion. Fluoroscopic image sequences (a series of 2-D x-ray projection images) can be used to assess real-time organ motion to help design an appropriate treatment margin, and to perform patient position monitoring during delivery. An important limitation of this x-ray technology is the inherent low contrast of soft-tissues. To overcome the problem of the target’s low visibility in images, many clinics insert several fiducial markers in the close proximity to the tumor. We propose a method for robust automatic localization of markers in low quality fluoroscopic images.

    • A daily 3D cone-beam CT image is produced with a patient in treatment position to verify the target/markers position before each treatment fraction. The acquisition takes about 2 minutes and hence cannot be used to characterize organ motion that happens during the treatment fraction. Fluoroscopy, on the contrary, provides sufficient means for motion tracking with 5.5 or more frames per second acquisition but the images are much harder to analyze.

    • We have developed algorithms and software for robust automatic marker localization and tracking, where localization is the most challenging part due to the small size of the markers and their very local intensity contrast. We use the cone-beam CT image and knowledge of the system’s geometry to predict the fiducial marker positions in the fluoroscopic images, and then correct the prediction by registering a template generated from the cone-beam CT to a pre-processed fluoroscopic image. Pre-processing and marker prediction are crucial in reducing computational cost and producing robust solutions.

    • In this presentation, the algorithmic details on digital filtering, in particular, a novel marker enhancement filter, as well as template generation methodologies and image registration followed by the validation approaches will be discussed. The described algorithms are not limited to the fluoroscopic tracking for organ motion assessment and can successfully be adapted to other applications.

    • Date: Monday, June 13th, 2011, 2-3pm

    • Location: Princess Margaret Hospital, Room 7-605,610 University Avenue, Toronto, Ontario (map)

    • Speaker: Paul Dufort, PhD, Computational Imaging Scientist at the Joint Department of Medical Imaging

    • Mount Sinai Hospital, University Health Network, Women's College Hospital

    • Title:

    • Joint Segmentation and Deformable Registration of Fractured Vertebrae Using a Synthesis of the Expectation Maximization and Belief Propagation Algorithms

    • Abstract:

    • A new surgical procedure aims to restore the height of damaged vertebral bodies that have suffered a compression fracture as a result of severe trauma. A computational method is needed to assess the efficacy of the procedure by automatically measuring the change in shape of the vertebral bodies in CT scans taken before and after the procedure.

    • While nominally a problem of registering the CT scans, the details of the required measurements pose unique difficulties. Rigid registration alone is inadequate due to the differing curvatures of the spine across scans, and the fact that the target vertebra may itself have changed shape substantially due to the surgical procedure. While the shape changes of interest could be derived from a deformable registration field, intensity-based registration techniques in general are problematic due to the large quantity of extremely electron-dense (3000 HU) materials injected into the vertebra during the surgical procedure, greatly reducing the image quality.

    • Instead, we describe a joint segmentation/registration technique based on deformably registering a model vertebra to the target vertebra in the CT scans before and after the procedure. While the vertebral body is damaged in these cases, the posterior regions are typically left intact. The segmentation allows the posterior regions of the vertebra to be isolated and rigidly registered, so that the remaining misalignment of the vertebral body can be attributed to physical changes in its shape.

    • The joint segmentation/registration method is feature-based, attempting to match a dense set of simple features extracted from the model vertebra surface to features extracted from the CT data. The expectation-maximization (EM) algorithm is used to simultaneously compute the deformation field and the match correspondence probabilities. However, since EM is an ascent-based mechanism, it is prone to converging to inappropriate minima in the presence of the clutter introduced by nearby adjacent vertebrae, ribs, aortic calcifications, and the cement and implants used in the procedure.

    • We therefore augment the expectation step with one or more loopy belief propagation (BP) iterations on an associated Markov random field that assign low probabilities to matches less consistent with the global shape of a vertebra. A probability metric is described for quantifying the agreement between two shapes that is invariant to translation and rotation, and whose sensitivity to scaling can be adjusted independently of shape. We demonstrate the technique and discuss its application, pros, and cons in the general context of finding correspondences in medical image data.

    • Date: Monday, May 2nd, 2011, 2-3pm

    • Location: Princess Margaret Hospital, Auditorium 6-604, 610 University Avenue, Toronto, Ontario (map)

    • Speaker: Mike Daly, University of Toronto / Princess Margaret Hospital

    • Title:

    • Intraoperative CBCT and Virtual Endoscopy for Head & Neck Surgery

    • Abstract:

    • Head and neck cancer surgery presents the challenge of tumour resection in close proximity to critical structures such as the carotid arteries, optic nerves, and brain. The need for precise localization within complex 3D anatomy has motivated the development of an image-guidance system that accounts for tissue deformation and excision through the use of intraoperative cone-beam CT (CBCT) imaging and real-time virtual endoscopy.

    • The imaging system is based on a prototype mobile C-Arm for intraoperative CBCT that provides 3D images with sub-mm spatial resolution and soft-tissue visibility. Geometric registration of endoscopic video with CBCT is achieved by way of a real-time optical tracking system and camera calibration techniques. Endoscopy-CBCT registration enables virtual endoscopic views of semi-transparent CBCT surface renderings which can reveal the presence of segmented critical structures not directly visible in the real endoscopic view.

    • Pre-clinical studies demonstrate the benefits of integrated CBCT and virtual endoscopy for improved localization accuracy and reduced surgeon workload, and motivate translation into an ongoing CBCT-guided head and neck surgery clinical trial.

    • Date: Monday, April 4th, 2011, 2-3pm

    • Location: Princess Margaret Hospital, Auditorium 6-604, 610 University Avenue, Toronto, Ontario (map)

    • Speaker: Fernando Flores-Mangas, University of Toronto

    • Title:

    • Shape-based registration

    • Abstract:

    • Renal lesions are very common. Most of them are benign cysts but some are cell carcinomas. Identifying the type of lesion requires measuring the brightness difference between two CT scans, one taken prior to the injection of a contrast agent (pre-contrast), and another one taken after injection (post-contrast). Currently, evaluating the level of contrast enhancement across corresponding image locations is a very tedious task, requiring visual inspection and manual 3D point matching for each region of interest. In this talk I will present a shape-based registration method that determines the level of contrast enhancement for the entire kidney, and does so with minimum user intervention. The approach builds a patient-specific 3D shape model of the kidney using the post-contrast image. This model is then matched against candidate surfaces extracted from the pre-contrast image. Candidate surfaces are generated using a novel grouping and segmentation approach, driven by iteratively redefined curvature constraints. The algorithm reaches a solution when several surfaces agree on the location of the model.

    • Our technique overcomes the limitation of existing algorithms where all image structures contribute equally to the resulting registration transformation. Results from experiments with 20 datasets of real patients with renal lesions indicate that our method produces highly accurate (within voxel size) registration results of kidney structures.

    • Time: Monday, March 21th, 2011, 2-3 pm

    • Meeting CANCELED

    • Date: Monday, March 7th, 2011, 2-3 pm

    • Location: Princess Margaret Hospital, Auditorium 6-604, 610 University Avenue, Toronto, Ontario (map)

    • Speaker(s): Doron Dekel, Co-CEO and CTO, Claron Technology

    • Title:

    • Building a medical image processing business

    • Abstract:

    • Claron Technology, headquartered in downtown Toronto, built its business on its capacity to efficiently develop innovative medical image processing software. The company, who was listed as Canada's 82nd fastest growing company in 2010, has been profitable in each of the 10 years since it was founded, and now counts 30 employees, nearly all software engineers. Its technology is being used daily by clinicians in thousands of locations worldwide.

    • By telling stories of how the company overcame various technical and business challenges, the presentation will shed light on what it takes to convert innovative medical image processing ideas to actual products used by clinicians to help patients.

    • The audience will be encouraged to actively participate.

    • Date: Tuesday, February 22nd, 2011, 1-2pm

    • Location: Bahen Centre for Information Technology (BA), Room 1200, 40 St George Street, Toronto (map)

    • Speaker(s): Intro to the seminar series by David Jaffray, followed by a talk by Vladimir Pekar, Stephane Allaire, Arish Qazi

    • Title:

    • Image Processing in Radiation Therapy

    • Abstract:

    • Manual contouring of target volumes and organs at risk in image-guided radiation therapy is extremely time-consuming, where a single patient treatment plan can take several hours to contour. As radiation treatment delivery moves towards adaptive treatment, the need for more efficient segmentation techniques will increase.

      • In the first part of the presentation, we introduce an automated segmentation method for head and neck data using salient feature-based deformable registration. The method makes use of extraction and matching of salient interest points using a 3-D extension of the scale-invariant feature transform (SIFT) algorithm. We present qualitative and quantitative results for automated ROI propagation from the planning CT data to selected daily cone-beam CT fractions acquired during the radiation therapy course.

    • The second part of the presentation is devoted to a novel approach for probabilistic refinement of segmentations obtained using 3D deformable models. It assumes that an uncertainty area exists around the boundary defined by the deformable model and that subsequent classification of this region into object / non-object classes will improve the segmentation accuracy. This concept is implemented by the following steps. First, a probabilistic mask is created by averaging the registered expert segmentations. This mask is next registered with the result of model-based segmentation in a new patient and the uncertainty area is refined using voxel classification based on a plurality of low-level image features. Performance of the method is compared to the conventional model-based scheme by segmentation of three important organs at risk in the head and neck region: mandible, brainstem, and parotid glands.