Optimize the system accuracy by leveraging AI The whole system needs to be precisely calibrated ✓ Acquisition parameters must be optimized March 20th, 2019 25/59
Optimize the system accuracy by leveraging AI The whole system needs to be precisely calibrated ✓ Acquisition parameters must be optimized ✓ System must be calibrated geometrically and temporally March 20th, 2019 25/59
Optimize the system accuracy by leveraging AI Tracking camera Ultrasound system Temporal delay The whole system needs to be precisely calibrated between images and position information ✓ Acquisition parameters must be optimized How to synchronize hronize them em ? ✓ System must be calibrated geometrically and temporally Images Position March 20th, 2019 25/59
Optimize the system accuracy by leveraging AI The whole system needs to be precisely calibrated ✓ Acquisition parameters must be optimized ✓ System must be calibrated geometrically and temporally March 20th, 2019 25/59
Optimize the system accuracy by leveraging AI The whole system needs to be precisely calibrated ✓ Acquisition parameters must be optimized ✓ System must be calibrated geometrically and temporally ✓ Speed of sound must be compensated March 20th, 2019 25/59
Optimize the system accuracy by leveraging AI The whole system needs to be precisely calibrated US systems assume a constant speed of sound However, sound travels at different speeds in fat and muscle ✓ Acquisition parameters must be optimized ✓ System must be calibrated geometrically and temporally US System Assumption 1540 m/s ✓ Speed of sound must be compensated Fat 1470 m/s Muscle 1620 m/s March 20th, 2019 25/59
Optimize the system accuracy by leveraging AI The whole system needs to be precisely calibrated ✓ Acquisition parameters must be optimized ✓ System must be calibrated geometrically and temporally ✓ Speed of sound must be compensated March 20th, 2019 25/59
Optimize the system accuracy by leveraging AI The whole system needs to be precisely calibrated ✓ Acquisition parameters must be optimized ✓ System must be calibrated geometrically and temporally ✓ Speed of sound must be compensated Such processes are usually tedious and complex March 20th, 2019 25/59
Optimize the system accuracy by leveraging AI The whole system needs to be precisely calibrated ✓ Acquisition parameters must be optimized ✓ System must be calibrated geometrically and temporally ✓ Speed of sound must be compensated Such processes are usually tedious and complex ...but we can leverage our real-time algorithms to solve them! March 20th, 2019 25/59
1) Parameter Tuning - Auto-Focus for Cameras Cameras can automatically find the region of interest in an image and optimize the acquisition parameters Focus Exposure Time March 20th, 2019 26/59
1) Parameter Tuning - Auto-Focus for Cameras Cameras can automatically find the region of interest in an image and optimize the acquisition parameters Focus Exposure Time March 20th, 2019 26/59
1) Parameter Tuning - Auto-Focus for Cameras Cameras can automatically find the region of interest in an image and optimize the acquisition parameters Focus Exposure Time March 20th, 2019 26/59
1) Parameter Tuning - Auto-Focus for Ultrasound! Frequency Brightness Focus March 20th, 2019 27/59
1) Parameter Tuning - Auto-Focus for Ultrasound! Frequency Brightness Focus March 20th, 2019 27/59
1) Parameter Tuning - Auto-Focus for Ultrasound! Frequency Brightness Focus March 20th, 2019 27/59
1) Automatic Acquisition Parameter Tuning March 20th, 2019 28/59
1) Automatic Acquisition Parameter Tuning Focus is equal to the • depth of the bone March 20th, 2019 28/59
1) Automatic Acquisition Parameter Tuning Focus is equal to the • depth of the bone Frequency also depends • on the depth of the bone (high frequencies do not travel deep enough) March 20th, 2019 28/59
1) Automatic Acquisition Parameter Tuning Focus is equal to the • depth of the bone Frequency also depends • on the depth of the bone (high frequencies do not travel deep enough) Brightness • can be adjusted by computing intensity statistics March 20th, 2019 28/59
L IVE D EMO A UTO -F OCUS in partnership with March 20th, 2019 29
2) Calibrations Speed of sound correction Temporal calibration Salehi & Prevost et al. Precis ise e Ultra rasoun und d Bone Regis istrat tratio ion n with Learning ning-Based sed Segment entati tion n and Speed ed of Sound d Calibr brati tion MICCAI 2017 March 20th, 2019 30/59
P ART 3 N EURO S URGERY March 20th, 2019 31
From planning to brain surgery • Brain surgery usually planned on pre-operative MRI Where is the tumor? How big is it? March 20th, 2019 32/59
From planning to brain surgery • Brain surgery usually planned on pre-operative MRI Where is the tumor? How big is it? • In the OR, very difficult to follow a surgical plan • Brain n shif ift: t: When the skull is opened, gravity causes the brain to collapse Lu, Jun-Feng, et al. NeuroImage: Clinical 2 (2013): 132-142 March 20th, 2019 32/59
From planning to brain surgery • Brain surgery usually planned on pre-operative MRI Where is the tumor? How big is it? • In the OR, very difficult to follow a surgical plan • Brain n shif ift: t: When the skull is opened, gravity causes the brain to collapse • Idea: Acquire ultrasound during surgery Deformable registration to the MR image → Planning can be used Lu, Jun-Feng, et al. NeuroImage: Clinical 2 (2013): 132-142 March 20th, 2019 32/59
From planning to brain surgery • Brain surgery usually planned on pre-operative MRI Where is the tumor? How big is it? • In the OR, very difficult to follow a surgical plan • Brain n shif ift: t: When the skull is opened, gravity causes the brain to collapse • Idea: Acquire ultrasound during surgery Deformable registration to the MR image → Planning can be used Lu, Jun-Feng, et al. NeuroImage: Clinical 2 (2013): 132-142 March 20th, 2019 32/59
MRI to 3D Ultrasound Registration March 20th, 2019 33/59
MRI to 3D Ultrasound Registration March 20th, 2019 33/59
MICCAI CuRIOUS Challenge 2018 Correction of Brainshift with Intra-Operative Ultrasound https://curious2018.grand-challenge.org March 20th, 2019 34/59
Not an AI method ! Wein et al. Global Regis istr tratio ation n of Ultra rasound und to MRI Using g the LC2 Metri ric for Enabli ling ng Neuro rosurg urgic ical al Guidanc nce MICCAI 2013 … but still computationally intensive → GPU implementation Top 3 methods were not based on machine learning March 20th, 2019 35/59
P ART 4 U LTRASOUND F OR V ASCULAR I MAGING in partnership with www.pi .piurimag rimaging ing.com .com March 20th, 2019 36
Vascular Imaging • Visualization of blood vessels • Multiple clinical applications, e.g. • Stenosis/Aneurysm Management and Surveillance • Fistula Planning and Monitoring • Vascular Mapping • Typically performed with a CT or MR scanner after injection of contrast agents → Expensive, long, toxic MR → Not suited for screening or monitoring March 20th, 2019 37/59
Vascular Imaging • Visualization of blood vessels • Multiple clinical applications, e.g. • Stenosis/Aneurysm Management and Surveillance • Fistula Planning and Monitoring • Vascular Mapping • Typically performed with a CT or MR scanner after injection of contrast agents → Expensive, long, toxic MR US → Not suited for screening or monitoring source: piurimaging.com March 20th, 2019 37/59
From 2D to 3D US – without External Hardware Motori rized d Trans nsdu ducer cer Matrix ix Array ay Track ckin ing g “wobbler” “3d probe” (opt ptic ical/ al/EM) M) Existi sting ng Hardware Solutions lutions Philips xMatrix Limited field of view Limited field of view Expensive Temporal artifacts Decreased image quality Not portable Imag age-Ba Base sed Our Goal Reconstru structio ction No hardware March 20th, 2019 38/59
From 2D to 3D US – without External Hardware Motori rized d Trans nsdu ducer cer Matrix ix Array ay Track ckin ing g “wobbler” “3d probe” (opt ptic ical/ al/EM) M) Existi sting ng Hardware Solutions lutions Philips xMatrix Limited field of view Limited field of view Expensive Temporal artifacts Decreased image quality Not portable Imag age-Ba Base sed Our Goal Reconstru structio ction No hardware March 20th, 2019 38/59
From 2D to 3D US – without External Hardware Motori rized d Trans nsdu ducer cer Matrix ix Array ay Track ckin ing g “wobbler” “3d probe” (opt ptic ical/ al/EM) M) Existi sting ng Hardware Solutions lutions Philips xMatrix Limited field of view Limited field of view Expensive Temporal artifacts Decreased image quality Not portable Imag age-Ba Base sed Our Goal Reconstru structio ction No hardware March 20th, 2019 38/59
From 2D to 3D US – without External Hardware Motori rized d Trans nsdu ducer cer Matrix ix Array ay Track ckin ing g “wobbler” “3d probe” (opt ptic ical/ al/EM) M) Existi sting ng Hardware Solutions lutions Philips xMatrix Limited field of view Limited field of view Expensive Temporal artifacts Decreased image quality Not portable Imag age-Ba Base sed Our Goal Reconstru structio ction No hardware March 20th, 2019 38/59
From 2D to 3D US – without External Hardware Motori rized d Trans nsdu ducer cer Matrix ix Array ay Track ckin ing g “wobbler” “3d probe” (opt ptic ical/ al/EM) M) Existi sting ng Hardware Solutions lutions Philips xMatrix Limited field of view Limited field of view Expensive Temporal artifacts Decreased image quality Not portable Imag age-Ba Base sed Our Goal Reconstru structio ction No hardware March 20th, 2019 38/59
Image-based Motion Reconstruction Frame-to-frame motion estimation I 1 I 2 T 1 → 2 Algorithm T 1 → 2 I 2 Rigid Transformation I 1 March 20th, 2019 39/59
Recommend
More recommend
