2011_12_12_Slides_video

Views:
 
Category: Entertainment
     
 

Presentation Description

No description available.

Comments

Presentation Transcript

A new algorithm to substantially speed up MRI scans :

A new algorithm to substantially speed up MRI scans Faster scans can reduce clinical scan times from 45 to 15 minutes Berkin Bilgic PhD Candidate MRI Group, MIT

Speeding up MRI scans is crucial:

Speeding up MRI scans is crucial Magnetic Resonance Imaging (MRI) is a non-invasive method for imaging of human tissues MRI is a crucial tool for medical diagnosis and monitoring the progress of treatment 3

Speeding up MRI scans is crucial:

Speeding up MRI scans is crucial Magnetic Resonance Imaging (MRI) is a non-invasive method for imaging of human tissues MRI is a crucial tool for medical diagnosis and monitoring the progress of treatment However, MRI scans often take more than half an hour, which is uncomfortable in a closed and noisy environment Subjects are also required to keep perfectly still as motion severely hampers image quality 4

Speeding up MRI scans is crucial:

Speeding up MRI scans is crucial Magnetic Resonance Imaging (MRI) is a non-invasive method for imaging of human tissues MRI is a crucial tool for medical diagnosis and monitoring the progress of treatment However, MRI scans often take more than half an hour, which is uncomfortable in a closed and noisy environment Subjects are also required to keep perfectly still as motion severely hampers image quality It is crucial to reduce the scan times, especially for vulnerable subjects and pediatric imaging, without sacrificing from image quality 5

Multi-contrast data acquisition:

Multi-contrast data acquisition In clinical MRI, it is common to image the same region of interest under multiple contrast settings Each contrast is capable to emphasizing certain tissue types, e.g. healthy tissue vs. pathology This way, multi-contrast images increase diagnostic power of MRI 6

Multi-contrast data acquisition:

Multi-contrast data acquisition In clinical MRI, it is common to image the same region of interest under multiple contrast settings Each contrast is capable to emphasizing certain tissue types, e.g. healthy tissue vs. pathology This way, multi-contrast images increase diagnostic power of MRI 7 proton density

Multi-contrast data acquisition:

Multi-contrast data acquisition In clinical MRI, it is common to image the same region of interest under multiple contrast settings Each contrast is capable to emphasizing certain tissue types, e.g. healthy tissue vs. pathology This way, multi-contrast images increase diagnostic power of MRI 8 proton density T2 weighted

Multi-contrast data acquisition:

Multi-contrast data acquisition In clinical MRI, it is common to image the same region of interest under multiple contrast settings Each contrast is capable to emphasizing certain tissue types, e.g. healthy tissue vs. pathology This way, multi-contrast images increase diagnostic power of MRI 9 proton density T2 weighted T1 weighted

Data acquisition in MRI:

Data acquisition in MRI In MRI, we do not directly measure the images, but we sample their Fourier transform 10 acquired data

Data acquisition in MRI:

Data acquisition in MRI In MRI, we do not directly measure the images, but we sample their Fourier transform 11 acquired data MR image linear transform

Data acquisition in MRI:

Data acquisition in MRI In MRI, we do not directly measure the images, but we sample their Fourier transform The more samples we collect, the longer the scan takes. So we would like to collect less samples but still obtain high quality images 12 acquired data MR image linear transform

Data acquisition in MRI:

Data acquisition in MRI In MRI, we do not directly measure the images, but we sample their Fourier transform The more samples we collect, the longer the scan takes. So we would like to collect less samples but still obtain high quality images 13 25% of data image with artifacts linear transform 4 times reduced scan time

Nonlinear image reconstruction:

Nonlinear image reconstruction 14 25% of data nonlinear reconstruction 4 times reduced scan time Lustig et al. [1] Lustig et al . MRM 2007 By using a nonlinear processing technique called Compressed Sensing, Lustig et al. 1 obtained much better image quality than the simple linear transformation

Nonlinear image reconstruction:

Nonlinear image reconstruction By using a nonlinear processing technique called Compressed Sensing, Lustig et al. 1 obtained much better image quality than the simple linear transformation We exploit the similarity of tissue boundaries across multi-contrast images, and obtain still better results than Lustig et al. 15 25% of data 4 times reduced scan time Lustig et al. [1] Lustig et al . MRM 2007 nonlinear reconstruction

PowerPoint Presentation:

16 using 100 % of data linear transform Error: 0 % RMSE SRI24 Atlas 0.7 0

PowerPoint Presentation:

17 Lustig et al. 1 Error: 9.4 % RMSE using 25 % of data 4-times acceleration [1] Lustig et al . MRM 2007 9.4 % Lustig et al . 0.7 0 0.15 0

PowerPoint Presentation:

18 Our method Error: 2.3 % RMSE 9.4 % Lustig et al . 0.7 0 0.15 0 Joint Bayes 2.3 % using 25 % of data 4-times acceleration

PowerPoint Presentation:

19 using 100 % of data linear transform Error: 0 % RMSE TSE Scans : in vivo acquisition 0.6 0

PowerPoint Presentation:

20 Lustig et al. 1 [1] Lustig et al . MRM 2007 Error: 9.4 % RMSE 9.4 % Lustig et al . 0.1 0 0.6 0 using 40 % of data 2.5-times acceleration

PowerPoint Presentation:

21 Error: 3.6 % RMSE 9.4 % Lustig et al . 0.1 0 0.6 0 Joint Bayes 3.6 % Our joint reconstruction using 40 % of data 2.5-times acceleration

Impact and Limitations:

We showed up to 4-times less reconstruction error relative to the state of the art algorithm It is possible to substantially accelerate clinical scans while retaining high image quality 22 Impact and Limitations

Impact and Limitations:

We showed up to 4-times less reconstruction error relative to the state of the art algorithm It is possible to substantially accelerate clinical scans while retaining high image quality Our initial implementation suffers from long processing times on the order of hours . We are expecting significant speed up from exporting our method to Graphics Processing Card (GPU) platform 23 Impact and Limitations

authorStream Live Help