Description: When choosing imaging, image processing, displacement post-processing, and strain calculation parameters, it is important to have a standard to compare to optimize these settings. For this purpose, we use the displacement and strain error summary results stored in the variable aSummaryALL for a correlation error analysis of an image set with average quality, exported by the Matlab functions:

process_results_correlationErrorStats3Strain.m

process_results_BaselineError.m

Baseline Error - Comparing two consecutive image volumes taken under the same conditions. Any displacements between the two volumes and the strains they result in are referred to as the Baseline Imaging errors. These are the practical limit for displacement and strain resolution as any changes smaller than this may just be due to baseline variations. We use this error to study the effect of imaging settings and image processing settings and to improve them. Example: If baseline displacement errors are high we might reduce the imaging time to reduce the influence of creep. We want average positional errors to be less than half a pixel ideally. If baseline strain errors are high, perhaps the strain calculation method can include pre-smoothing to reduce errors due to baseline displacement variations.

DVC Correlation Error - Applying a rigid body displacement and strain in 3D to the reference image volume, correlating with DVC, and calculating the difference between the applied and calculated field. This analysis tells us where DVC can accurately detect motion and where it can't (due to poor contrast or other issues). This kind of error is used to optimize the correlation error threshold, to optimize the error threshold used to mask out areas of poor correlation based on this analysis, and to improve the strain calculation and outlier removal methods.

Lastly, we calculate 4 metrics per displacement or strain error component: average error (bias), standard deviation, average absolute error (measure of average local error), and absolute standard deviation. A high bias will tell you that there is a large, systemic error in the DVC calculation. A high absolute average error will tell you the expected average local error, which is a better estimate of accuracy in any particular area.

Please refer to the attached bundle of materials for examples of this process and protocols for optimizing each kind of setting used in DVC analysis and in imaging for current projects.

protocol_for_dvc_displacement_and_strain_error_study.docx powerpoint_example_of_displacement_processing_study.pptx mouse_lc_error_summary.xlsx