duke nirl hosted miriad data both xnat and hid
TRANSCRIPT
Multispectral Segmentation 1.5T and 3T
1.5T Segmentation 3T Segmentation
Source scans: proton density, T1-w, T2-w, FLAIRAutomated segmentation of gray, white matter, csf and lesion (red). Shown overlaid on proton density image (showing just skull, surface tissues and reference tube outside skull)
500
550
600
650
700
750
gm 3
500 550 600 650 700 750
gm 1.5
0
5
10
15
20
25
30
35
les
3
0 5 10 15 20 25 30 35
les 1.5
Lesion SimulationFractional Contrast for Lesion to White Matter, 3T data
PD-weighted
T1 weighted T2-weighted FLAIR
Mean 0.299168 -0.12935 0.88210114 0.356158
Standard Deviation (SD) of Mean 0.041301 0.020395 0.13435156 0.097343
SD/Mean 0.138053 -0.15767 0.152308567 0.273315
VSL=volume of segmented lesion inside of true lesion (True positives) VNSL=volume of nonsegmented lesion inside of true lesion (False negatives)
VNL=volume of non-lesion outside of true lesion (True negatives) VLOL=volume of segmented lesion outside of true lesion (False positives)
sensitivity=VSL/(VSL+VNSL) specificity=VNL/(VNL+VLOL) accuracy=VSL/(VSL+VLOL)
For this example we have sensitivity = 0.73, specificity = 0.89 and accuracy=0.46.
SENSE DTI Image Distortion Correction
B0 and eddycurrent maps
distortedSENSE DTI
images
correctedSENSE DTI
imagesdifference
Toward Quantitative Fiber Volume Measurement
Left Column: Conventional Euler’s methodRight Column: Improved streamline tracking method with RK4 algorithm.
a b
c d
a b
c d