Download presentation
Presentation is loading. Please wait.
1
McDaniels – Oct 10, 2008
2
Outline Geometric Uncertainty Uncertainty in average intensity due to lesion placement ADC uncertainty
3
Geometric Uncertainty Uncertainty in number of voxels – Transcription from contour to DW image, 4 contour voxels for each DW voxel – Algorithm rounds to nearest voxel – Voxels in center of lesion certain – Voxels on perimeter of lesion uncertain – Estimate each DW voxel on average has +/- ¼ – Affects weighted calculation of whole lesion
4
Geometric uncertainty Lesion geometry, roughly circular section Area => n = πr 2 Perimeter => n p = 2πr +/- ¼*(2πr ) σ n = πr/2 = √(n/ π) For n = 150, σ n ≈ 7 Typical uncertainty about 2-3% Doesn’t include 2-3% scaling from Contour to DW
5
Geometric uncertainty Slice matching – Contour – 4mm slice, 4mm spacing – DW – 5mm slice, 7mm spacing Not all contour slices have a matching DW slice All DW slices have matching contour slice DW slice overlaps contour slice
6
Lesion Placement Estimated by placing ROI and analyzing histogram, then shifting and reanalyzing Variation in average intensity for shifts of 1-2 pixels was about 2-3%
7
Uncertainty in ADC Values Statistical uncertainty in image intensity – Statistical fluctuation – Structure (GM, WM, CSF) Typically 15-30% -ln(I/Io) => σ = √(0.15 2 +0.15 2 ) ≈ 0.21 Linear fit from fixed point and two points with uncertainty
8
Uncertainty in ADC Values Uncertainty in linear fit from Bevington σ 2 = 1/Δ*Σ1/σ i 2 Δ = Σ 1/σ i 2 * Σx i 2 / σ i 2 -(Σ x i / σ i 2 ) 2 e.g. for σ 2 =σ 3 =.2, x 2 =520, x 3 =850 Δ = (1/.04 +1/.04)*(520 2 /.04+850 2 /.04)- (520/.04+850/.04) 2 = 6.8e7 σ = √(1/6.8e7*(1/.04+1/.04)) = 0.00086 ADC ≈ 0.0015 +/- 0.00086 (57%)
9
ADC uncertainty Patient 5 Minimum +/-43% Maximum +/-215% Average +/-104%
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.