Making it the Biomedical Way The CMI Cardiac
-- Making it the Biomedical Way The CMI Cardiac Action Potential Imaging System An Arrhythmia Research Tool BC Research Institute for Children’s and Women’s Health Simon Fraser University
Agenda • Introduction • Background • Technical Features & Future Improvements • Business Component • Conclusion • Q&A
Introduction – System Overview Optical Pathway Imaging Device Image Processing Image Generation Image Acquisition Image Analysis
Introduction – CMI Executive Team Seddrak Luu President Ronnie Chan CEO Deanna Lee COO Edwin Wong CTO Hardware Yindar Chuo CFO Jimmy Tsui CTO Firmware Allen Lai CMO Stephen Wong CTO Software
Introduction – Sources of Information Dr. Andrew Rawicz Dr. Glen Tibbits Haruyo Kashihara Eric Lin School of Kinesiology School of Engineering Science Dr. M. Faisal Beg
Background – Arrhythmia Junctional Ectopic Tachycardia • Disorders of the regular rhythmic beating of the heart • Occur in children after open heart surgery • 1 in 10 infants goes into arrhythmia after open heart surgery • Can be lethal
Background – Our Goal • Develop CAPIS to help understand the arrhythmia • Focus on helping those in need
-- Making it the Biomedical Way Technical Features of CAPIS
-- Making it the Biomedical Way Technical Features of CAPIS • Image Generation Module • Image Acquisition Module • Image Analysis Module
Image Generation Module – Setup • Mercury Arc Lamp and Blue LED • Rabbit Heart cell Heart solution • Mirrors and filters in multiviewer in microscope • Stimulator • Photomultiplier Tubes • Lab. VIEW software Original Current Setup
Image Generation Module – Cell Preparation • Heart cell extraction from 20 -day-old rabbit • Addition of dye and calcium • Pipette solution into bath Bath • Connect wires from bath to stimulator Bath secured on holder
Image Generation Module – Optical Pathway • Cells are exposed to filtered light source • Stimulator is turned on • Cell contracts, emits light from fluorescence • PMT captures green/red photons in form of TTL pulses
Image Generation Module – Cell Contraction Finding Healthy Contracting Cells • Cell should be elongated (as opposed to round) • Cell should have striations • Neighbouring cells should not crowd or overlap contracting cell to optimize signal-to-noise ratio when capturing its fluorescence • Cell contracts by shrinking approx. 15% at each end
Image Generation Module – Cell Contraction
Image Generation Module – Results Mercury Arc Lamp • Cell receives correctly filtered excitation beam • Cell contracts and fluoresces under stimulation and excitation Blue LED • Cell receives correctly filtered excitation beam • Not enough power to observe contractions
Image Generation Module – Results
Image Generation Module – Results Software • TTL pulses from PMT can be counted by software • Software can output ratiometric info to text file • Software can measure frequency, but tradeoff exists between sensitivity and accuracy • Ratiometric data collected by software does not agree with anticipated results
Image Generation Module – Results Possible causes for results not meeting expectations • Found very few healthy cells, and most of those found to be overlapped, causing poor signal-tonoise ratio • Time constraint to conduct experiments and collecting data • Experimental techniques may be a factor • Software may lack accuracy in producing our expected results
Image Generation Module – Feasibility Further steps necessary to ensure feasibility • Additional observations of healthy cells • Ensure the software produces results accurate enough to meet our needs
-- Making it the Biomedical Way Technical Features of CAPIS • Image Generation Module • Image Acquisition Module • Image Analysis Module
Image Acquisition Module – Development Overview • Module goal for proof-of-concept phase • Design and functionality overview and verification • Performance analysis for benchmarking • Recommendations drawn from analysis • Future considerations
Image Acquisition Module – Goals (1) • The concept being proved – Image capture and display • Approach to verifying the concept • Available resources
Image Acquisition Module – Goals (2)
Image Acquisition Module – Functionality Overview • Input requirement • Physical requirement • Image capturing performance requirement • Image display performance requirement • General requirement
Image Acquisition Module – Design Overview (1) Devices and software provided: 1. Prosilica CV 640 Machine Vision Camera (CMOS technology, IEEE 1394 A Fire. Wire connection) 2. Laboratory computer with monitor 3. National Instruments’ Lab. VIEW Graphical Development Environment 4. National Instruments’ Advanced IMAQ Vision for Lab. VIEW 5. NI-IMAQ IEEE 1394 Machine Vision Support Package
Image Acquisition Module – Design Overview (2)
Image Acquisition Module – Design Overview (3)
Image Acquisition Module – Design Overview (4) Image capturing
Image Acquisition Module – Design Overview (5) Image control • Prosilica camera driver • NI IMAQ for IEEE 1394 driver
Image Acquisition Module – Design Overview (6) Graphical user interface • NI Lab. VIEW and IMAQ develop environment • Program flow • GUI operation and functionalities (demo)
Image Acquisition Module – Functionality Validation Confirm camera operations • Continuous image • Camera feature controls Confirm Lab. VIEW GUI operations • Display, save, open, pseudo-colour, and digital zoom for still and continuous image • Additional features
Image Acquisition Module – Performance Analysis (1) Frame Rate • Test procedure • Results § Sample Image Acquisition Module video test capture § Sample consumer camera test capture • Conclusions drawn
Image Acquisition Module – Performance Analysis (2) Intensity • Test procedure • Results § § § • Rabbit heart specimen Image under microscope 10 x at half 6 V 30 W Top: shutter = 4095, brightness = 255, gain = 0 Bottom: shutter = 4095, brightness = 255, gain = 16 Conclusion drawn
Image Acquisition Module – Performance Analysis (3) Wavelength • Test procedure • Results § Top Left: Multi-wavelength image captured by IAq. M at Prosilica camera setting of shutter = 750, gain = 0, brightness = 0, gamma = 0 § Top Right: Multi-wavelength image captured by KODAKLS 443 § Bottom: Quantum efficiency graph of the Prosilica CV 640 CMOS digital camera • Conclusion drawn
Image Acquisition Module – Performance Analysis (4) Resolution • Test procedure • Results § Side: Rabbit heart cell image captured by IAq. M via microscope at 10 x overall magnification § Field of view and clarity • Conclusion drawn
Image Acquisition Module – Next Steps and Future Improvements • Overview design for functional prototype • Hardware performance improvements • GUI output and control improvements
-- Making it the Biomedical Way Technical Features of CAPIS • Image Generation Module • Image Acquisition Module • Image Analysis Module
Image Analysis Module: Overview 1. 2. 3. 4. 5. 6. 7. 8. Problem Definition & Module Goals Module Specifications Our Approach System Design: Geodesic Splines System Design: MATLAB LWM Transform Test Data Sets Test Results Recommendations for Future Development
Problem Definition: Motion Artifact • Two types of motion artifact: 1. Intra-frame blurriness – – Cause: undersampling Require high frame rate from CCD camera 2. Inter-frame variations – – – Image changes shape from frame to frame Difficult to analyze Requires image registration This is where we come in!
Module Goals • • Match input image with base image Boundary matching Interior stretching Preserve gradient
Module Specifications • General requirements – • Hardware requirements – • Reasonable processing speed and efficiency Input requirements – – – • Post-capture processing, done on a per-frame basis 8 -bit, grayscale polygons Gradient images User-defined landmarks Performance requirements – Save and display results
Our Approach • • • Landmark-based image registration COTS vs. Customized Research Product Therefore, two branches: 1. Geodesic Splines – Based on research of Dr. Faisal Beg, Medical Image Analysis Lab (MIAL), SFU 2. MATLAB (software by The Mathworks, Inc. ) – Using in-the-box image registration capabilities
Younes and Camion Geodesic Splines Optimisation • General description • Minimising energy based on both smoothness and data fitting term • Iterative process called “gradient descent” • Implemented by Dr. Faisal Beg
Block Representation Windows or Linux
Input Parameters σ=4 • Are application specific – Depend on inputs (landmarks) and the goal of the matching • Major Parameters – ε-Energy – Green’s kernel – λ weighting σ=8 σ = 16
System Design: MATLAB Branch • • MATLAB 6 Release 12 application Local-weighted mean (LWM) transform Command line interface Research grade software – – Focus: evaluate algorithm performance Not as robust as it can be
System Flow Chart
Test Data 1: Demonstrative Polygons • • Grayscale, gradient polygons Generated in Adobe Photoshop Successive transformations Simulate shape variations in series of images
Test Data 2: Fluorescent Heart Images • • • Realistic fluorescent rabbit heart images Similar to expected input images Source: Living State Physics, Vanderbilt University www. vanderbilt. edu/lsp/panoraming. htm
Simulating Quantum Dots • What are quantum dots (QD)? – – • Semiconductor nanocrystals Fluorescent labels as landmarks Salt noise to simulate QD – – Distinguished labels Randomness Add noise Test image to test image Fluorescence Generate salt image noise with image QD +
Live Demo
Demonstrative Polygons Test Polygon Base Input 1 Input 2 Input 3
Demonstrative Polygons Test MATLAB Base All Interior Landmarks Only Input 1 to. Base Input 2 to. Base Input 3 to. Base Geodesic Splines Base
Demonstrative Polygons Test All Interior Landmarks Only M A T L A B G S Input 1 to. Base Input 2 to. Base Input 3 to. Base
Demonstrative Polygons Test Geodesic Spline Polygons: Input 01 to. Base Input 02 to. Base Input 03 to. Base All: 98. 480% 98. 466% 97. 614% Interior: 96. 904% 96. 451% 96. 961% MATLAB Polygons Input 01 to. Base Input 02 to. Base Input 03 to. Base All: 98. 171% 74. 314% 72. 598% Interior: 98. 364% 74. 314% 98. 042%
Fluorescence Image Test Input Image Base Image
Fluorescence Image Test All Interior Landmarks Only M A T L A B Input Image Base Image Input. To. Base G S
Fluorescence Image Test MATLAB All landmarks MATLAB Interior points Geodesic Spline All landmarks Geodesic Spline Interior points Geodesic Splines Fluorescence Image All: 84. 532% Interior: 83. 857% All: 90. 571% Interior: 77. 137% MATLAB Fluorescence Image
Current Limitations • • • MATLAB Image Registration – Cumbersome to select landmarks manually – Some landmarks fade after multiple iteration – Sensitive to landmark placement Geodesic Splines Image Matching – Computation time – Operating system constraint (temporary) Overall System Scheme – Non-practical for large sets of images
MATLAB vs. Geodesic Spline • • Geodesic Spline matching greater accuracy MATLAB is faster Short-term solution: MATLAB Long-term solution: Geodesic Splines
Recommendations for Future Development • • Automate/semi-automate landmark selection Increase efficiency in landmark-matching tools – – • Modify image I/O (e. g. Batch Mode operation) Input parameter optimization in geodesic splines method Automating geodesic splines matching pathway
-- Making it the Biomedical Way Business Component of CAPIS
Business Component – CMI Core Business 1. Cardiac Action Potential Imaging System (CAPIS) 2. Customize system to suit the need 3. Customize the Lab. VIEW Interface
Business Component - Client • Dr. Glen Tibbits of BC Research Institute of Children’s and Women’s Health
Business Component - Statistics • In British Columbia, ~40, 000 babies were born • 600 babies are born with congenital heart disease • Of that about 200 will require open heart surgery • 1 in 10 will go into Arrhythmia Source: Statistics Canada
Business Component – Marketing Strategies • Short Term – Research Based Institute – Children Hospitals and Children Research Institute • Long Term – Spin off into several groups • Customize software • Customize system – Selling to Research Institutes
Business Component – Proof of Concept – Research: $ 360. 35 • Apparatus – Company Registration/Administration: $512. 05 – Total Cost: $ 872. 40
Business Component – Product Price for Proof-of-Concept Component Price (CAD) 1. USB Fire. Wire cable $41. 81 2. Heat sink and accessories $18. 82 3. Bock Electronics (Lens for CCD Camera) $186. 61 4. Lexon blue LED and power supply $68. 11 NET PRODUCT PRICE (after tax) $362. 65
-- Making it the Biomedical Way Conclusion
Contact Information • • Website: Email: Phone: Address: www. sfu. ca/~sluu/concardio. html ensc-bcri@sfu. ca 604 -719 -5929 654 -8155 Park Road Richmond, British Columbia V 6 Y 1 S 9
-- Making it the Biomedical Way Q&A
- Slides: 71