3D Image Quality Metrics for Porosity in Tissue Scaffolds Craig Schroeder Advisors: Ana Ivelisse Aviles - Statistical Eng. Div., ITL Marcus Cicerone -

3D Image Quality Metrics for Porosity in Tissue Scaffolds

Craig Schroeder

Advisors:Ana Ivelisse Aviles - Statistical Eng. Div., ITL

Marcus Cicerone - ?? Div., MSEL

National Institute of Standards and Technology

August 11, 2005

Preliminaries

● Scaffold, tissue scaffold● Image (three-dimensional)● Porosity

– Volume of pore / total volume– 1 - density

Porosity Measurement

● Measure porosity– Scan in 3D scaffold

● MRI, CT scan

– Compute porosity

Image of MRI / CT equipment

Problem

● Inaccuracy in computed porosity– Scans contain noise– Noise affects computation– How accurate is the porosity?

Proposed Solution

● Develop an “Image Quality Metric”– Number computed for image– Higher quality indicates better accuracy

High Quality Low Quality

7 5 2 1

Metric Candidate One

● Find statistical relationships between– Computed porosity– Porosity after applying a filter– Actual porosity

● Estimate difference between– Computed porosity– Actual porosity

● “Image Quality Metric”

Stats

IQM

Original Noisy

Filtered

Statistical Relationships

Theory of Observed Slopes

● Clear trends in Least Squares angle● Explain trend with simple theory (Left)● Improve the theory for a better fit (Right)

Power Spectrum

● Signal processing● Amount of power broken down by frequency● Computed from Fourier transform

Metric Candidate Two

● Compute “power spectrum”● Find relationship between

– Computed porosity– Actual porosity– Power spectrum

● Use power spectrum to estimate error

Stats

IQM

Original Noisy

Power Spectrum

Noise Level vs Power Spectrum

● Need strong relationship between– Noise level (x axes)– Some aspect of the power spectrum (y axes)

Future Work

● Three weeks left!● Write this up as part of a publication● Consider permeability● Improve second metric

Image suggestsions?

Comments?

● Acknowledgments– Fredrick Phelan– Martin Chiang– NSF– NIST/ITL