Z. Rumboldt, S. Tipnis, D. Vicent,M.V. Spampinato, G. Goldsberry, W. Huda
Medical University of South CarolinaCharleston, SC, USA
XIX Symposium Neuroradiologicum
Iterative Reconstruction Algorithm for Head CT
BackgroundCT images traditionally reconstructed using filtered back projection techniques(FBP)FBP limitations: geometry, data completeness, radiation dose
Increased spatial resolution is directly
correlated with increased image noise
BackgroundITERATIVE IMAGE RECONSTRUCTION approaches recently proposed and introducedmay allow for improved image quality and lower noise(2 Alternative Forced Choice methodology)
iterative reconstruction may allow decoupling
of spatial resolution and image noise
need for a substantial increase in computation power
compared to conventional FBP reconstruction
After an image is reconstructed, “reprojection” simulates the CT measurement process, with the image as the object, followed by corrections
Each time the image is updated, processing algorithmenhances spatial resolution at higher object contrasts & reduces image noise in low contrast areas
Purpose
Novel Methodology
Comparison with normal scans for lesion detection, not side by
side
Clinical evaluation of potential noise reduction andimproved lesion detection
Material and MethodsPart 1
10 adult head CTs – both FBP and IRIS
2 neuroradiologists evaluated simultaneously both sets for each patient at 3 levels - MCP, BG, and centrum semiovale (30 levels in all) for noise and artifacts
raters blinded for the algorithm had 3 choices: preference for A, for B, no preference
Materials and Methods
Total screened 228 Age range selected 25 through 8524 abnormal subjects - 30 lesions: 21 hypo 7 hyper 1 mixed 1 iso
12 normal subjects selected to match
Part 2
Materials and MethodsIn house software (MUSC, Matthew Daniels, website accessible on campus network)
Displayed pairs of single slice CT - Abnormal on LeftLocation and description of lesion givenRating scale 1 to 10 1 = Barely discernable 10 = Definitely see lesion
PATHOLOGY COMPARED TO THE NORMAL 3 sets for each pair: FBP, NBC, IRIS
Materials and Methods
FBP NBCIRIS
FBP
IRIS
NBC
Every reader trained on practice set prior to study Individual randomization for every reader
AnalysisRatios of the obtained values: IRIS/FBP
NBC/FBP
Materials and MethodsEvery reader trained on practice set prior to studyIndividual randomization for every readerEvery reader trained on practice set prior to studyIndividual randomization for every reader
FBP IRIS
Results
Results
In all evaluated images (30 levels) noise was considered lower with IRIS compared to FBP
Artifacts were less prominent in 11 of the 30 evaluated levels using IRIS and in 3 using FBP (no preference was found for 16 levels)
Part 1
IRIS NBC
Results
Results
IRIS NBC
FBP (d) NBC (d') IRIS (d") d'/d d"/d1 3 1 3.00 1.001 3 3 3.00 3.008 7 7 0.88 0.886 3 2 0.50 0.336 2 5 0.33 0.834 7 6 1.75 1.509 6 7 0.67 0.786 7 3 1.17 0.503 3 5 1.00 1.673 3 5 1.00 1.671 2 4 2.00 4.006 7 3 1.17 0.503 4 5 1.33 1.677 1 5 0.14 0.714 7 7 1.75 1.757 5 9 0.71 1.293 5 7 1.67 2.338 8 8 1.00 1.002 4 6 2.00 3.004 4 7 1.00 1.752 4 2 2.00 1.002 7 8 3.50 4.004 6 7 1.50 1.753 8 6 2.67 2.003 7 8 2.33 2.677 6 10 0.86 1.435 5 3 1.00 0.603 4 7 1.33 2.338 7 10 0.88 1.254 6 4 1.50 1.00
Rater 1
Part 2
Rater 1 NBC/FBP IRIS/FBP
Avg. Hypo 1.34 1.48Std.Dev. Hypo 0.77 0.94
Avg. Hyper 1.51 1.61Std.Dev. Hyper 0.72 0.75
Average 1.45 1.61Std. Dev 0.82 0.97
Results
AVG. HYPO RATIO 1.34 1.48 1.69 1.49 1.07 1.22
SD HYPO RATIO 0.77 0.94 2.13 1.21 0.43 0.41
AVG. HYPER RATIO 1.51 1.61 1.12 1.18 1.02 1.05
SD HYPER RATIO 0.72 0.75 0.48 0.47 0.18 0.26
AVG. RATIO 1.45 1.61 1.54 1.42 1.09 1.24SD RATIO 0.82 0.97 1.82 1.04 0.41 0.45
NBC IRIS NBC IRIS NBC IRIS
Rater 1 Rater 2 Rater 3
Results Pooled Results NBC IRIS
Pooled Avg. Hypo 1.37 1.40Pooled Std.Dev. Hypo 1.34 0.91
Pooled Avg. Hyper 1.22 1.28Pooled Std.Dev. Hyper 0.53 0.56
Pooled Average 1.36 1.42Pooled Std. Dev 1.18 0.87
ConclusionIterative reconstruction algorithm decreases noise in Head CT images
It seems to improve lesion detection
It may allow decreased radiation dose
No clear difference between IRIS and NBC