Supporting Mental Health Awareness Week
Mental Health Awareness Week aims to educate about mental health issues in young people, highlighting the importance of early intervention and support. The week sheds light on the impact of mental illness on thinking, emotions, daily activities, and relationships. It emphasizes the significance of good mental health for young individuals to have control over their emotions, cognitive functions, and positive interactions. Additionally, the week addresses the challenges faced by children and youth with mental health difficulties, emphasizing the need for timely access to mental health services and support in schools.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
Estimating the Precision of Lesion Uptake Measurements M.E. Daube-Witherspoon, J. S. Karp, S. Matej, Y. Li, S.D. Metzler* Department of Radiology University of Pennsylvania Fully 3D 2017
Introduction Lesion quantification is affected by: Lesion characteristics (size, shape, local uptake) Clinical metric of interest (SUVmean, SUVpeak, SUVmax) Scan characteristics (spatial resolution, # events) Accuracy of corrections Reconstruction algorithm & parameters (# iterations, post- filtering, PSF modeling) Fully 3D 2017 2 iterations 10 iterations
Introduction Lesion quantification is affected by: Lesion characteristics (size, shape, local uptake) Clinical metric of interest (SUVmean, SUVpeak, SUVmax) Scan characteristics (spatial resolution, # events) Accuracy of corrections Reconstruction algorithm & parameters (# iterations, post- filtering, PSF modeling) Choice of optimal reconstruction for a given lesion dictated by both accuracy and precision of measurement Overall goal of project = Optimal reconstruction for a specific lesion in a given patient Fully 3D 2017
Estimating accuracy Using metrics less sensitive to partial volume errors SUVmax, SUVpeak Partial volume correction Based on simple phantom studies or Segmented image from anatomy & model of scanner Synthetically embedding multiple spheres into patient data Multiple locations, sizes Known uptake measurement of bias in a given patient Fully 3D 2017
Estimating precision Multiple independent realizations Easy in simulation; feasible in phantom studies Not practical for patient studies Precision depends on count density of study Synthetically embedding multiple spheres into patient data Multiple locations, sizes Statistically independent replicates embedded Bootstrapping Data sampled with replacement to generate multiple copies (same # cts) Useful to derive information about statistical properties of images Combination of embedding spheres bias correction and bootstrapping precision of VOI estimates I Buvat, Phys Med Biol 47: 1761-1775, 2002. M Dahlbom, IEEE Trans Nucl Sci 49: 2062-2066, 2002. C Lartizien & I Buvat, IEEE NSS/MIC, 2005. P Markiewicz, Phys Med Biol 60: 279-299, 2015. Fully 3D 2017
Simulations - 1 Cylinder with lung + 2 spheres (3:1 activity ratio) 100 independent realizations (10 M trues) simulated with EGS4 Up to 200 bootstrapped replicates each (10 M cts) from 10 different realizations precision of bootstrapped estimate 3 sphere sizes considered: 10, 22, and 37-mm diameter List-mode TOF OSEM reconstruction (5 iter., 25 subsets) Analysis: %SD across replicates of VOI values on spheres and background Diameter = physical diameter of sphere VOImean Maximum value in sphere VOImax Background VOI = 5-cm diam. circle Fully 3D 2017
Bootstrapping vs. Independent Realizations (10 Mcts & 100 replicates) (Error bars = SD of precision estimates across 10 starting points for bootstrapping) * For typical FDG count levels & 100 replicates, bootstrapping gives a good estimate of the precision of the VOI measurement. Fully 3D 2017
Simulations - 2 Question: Can we estimate lesion precision by embedding multiple realizations of spheres alone? Investigated components of data: 100 independent realizations of 22-mm spheres in 1 realization of background 2 different background realizations 100 independent realizations of background + 1 realization of 22-mm spheres 2 different sphere realizations Compared with 100 independent realizations of (background + spheres) Fully 3D 2017
Components of precision VOI 100 (S+B) 100 (S) + 1 B 1 S + 100 (B) Spheres 4.57% 2.61% 2.42% Bkgd 4.13% 3.50% 4.31% (22-mm spheres, independent realizations) Using only 1 replicate of the background or sphere data underestimates the precision of the spheres. To estimate precision of lesions in patient data, we will bootstrap the patient data + embedded spheres. Fully 3D 2017
Simulations - 3 10 Mcts = typical count density for clinical FDG study However, sphere events account for a small fraction of total phantom events: 37-mm: ~0.6% of total events 60 k sphere events 22-mm: ~0.15% of total events 15 k sphere events 10-mm: ~0.014% of total events 1.4 k sphere events As count density decreases, # sphere events in a given realization used for bootstrapping gets very small. Question: Does the good agreement between precision estimated using bootstrapping and independent realizations translate to lower count densities? Fully 3D 2017
Impact of count density 10 M 5 M 2.5 M 100 replicates * With 100 replicates, bootstrapping gives a good estimate of the precision of the VOI measurement for count densities down to 2.5 M true events. Fully 3D 2017
How many replicates are needed? (Error bars = SD of precision estimates across 10 starting points) For 10-Mcts study with 22-mm spheres, calculated %SD across 25, 50, 100, 150, and 200 bootstrapped replicates. Independent realizations only had 25, 50, and 100 replicates. Fully 3D 2017
How many replicates are needed? (Error bars = SD of precision estimates across 10 starting points) Using fewer than 100 replicates leads to increased variability in the measured precision for both VOImean and (especially) VOImax. Similar results were seen for lower count densities. Fully 3D 2017
Measurements NEMA image quality phantom (spheres 10-37 mm, 9.7:1 activity ratio) Philips Ingenuity TF scanner 25 independent realizations (5.75 M prompts, 3.35 M trues) 200 bootstrapped replicates from 5 different realizations precision of bootstrapped estimate Data corrections from 1 realization used for all replicates List-mode TOF OSEM reconstruction Analysis: %SD across replicates of VOI values on spheres and background Fully 3D 2017
Measurements Using fewer than 100 replicates leads to increased variability in the measured precision for all sphere sizes. Fully 3D 2017
Measurements Average estimate of %SD is stable down to ~50 bootstrapped copies. Some discrepancy observed between bootstrapping and independent realizations with measured data. Fully 3D 2017
Measurements Overall results for all sphere sizes are promising but warrant further investigation. Fully 3D 2017
Summary Simulation studies showed: Good agreement between precision estimates derived from bootstrapping and independent realizations for both VOImean and VOImax For 100 replicates For count densities as low as 2.5 M true events (22-mm sphere) Bootstrapping of (patient data + embedded spheres) will be needed to obtain an accurate estimate of the precision of sphere uptake Measured phantom data showed: Stable measurements of precision for 100 replicates Some inconsistency in precision estimates between bootstrapping and independent realizations, esp. for the smallest sphere Fully 3D 2017
Future work Study the impact of count density on 10-mm sphere 10-mm spheres contributed <2 k events to 10-Mct study Fewer sphere events may impact the precision of the uptake measurement More embedded spheres may be needed? Acquire phantom data with more independent realizations & more counts Incorporate bootstrapping of clinical patient data with embedded spheres into automatic expert system to determine the optimal reconstruction algorithm & parameters for a given lesion in a specific patient Fully 3D 2017
Acknowledgments Work was supported by the National Institutes of Health under grants #R21EB021559 and #5R01CA113941. Fully 3D 2017
