Assessment of 3D performance metrics. X-ray based Volumetric imaging systems: Fourier-based imaging metrics. The MTF in CT

Transcription

1 Assessment of 3D performance metrics D and 3D Metrics of Performance Towards Quality Index: Volumetric imaging systems X-ray based Volumetric imaging systems: CBCT/CT Tomosynthesis Samuel Richard and Ehsan Samei Carl E. Ravin Advanced Imaging Labs, Duke University Medical Center Fourier-based imaging metrics - Resolution NPS - Noise Detectability index observer performance Assessment of 3D performance metrics The in CT Challenges Measurement Interpretation NPS

2 NPS Pixel intensity The in CT The in CT Distance from center (re-binned) S. Richard et al. Towards a task-based approach of CT performance assessment: A novel measurement method of system and object Med Phys (submitted) Method comparison Noise-power spectrum (NPS) ACR phantom NPS= FT{ } 5 ma 5 ma ma ma um W wire Fourier transform of Noise field => NPS Measurement of noise and texture

3 Task based-optimization Task based-optimization... Task function Large feature..... NPS + AUC Mathematical Description of.5 imaging task..... NPS - - (mm) + Task function.5 5 Low frequencies AUC Task function weighs the and NPS by spatial frequencies relevant to imaging task. Task function weighs the and NPS by spatial frequencies relevant to imaging task. Task based-optimization NPS Small feature - - (mm) Higher frequencies AUC Model observers Fisher-Hotelling observer (FH)* Provides a theoretical estimate of observer SNR. Weighs the NEQ by frequencies relevant to imaging task Task function weighs the and NPS by spatial frequencies relevant to imaging task. *ICRU Report #5 3

4 AUC Model observers Fisher-Hotelling observer (FH)* Non-prewhitening Observer (NPW)** Does not compensate for correlation in the noise. Model observers Fisher-Hotelling observer (FH)* Non-prewhitening Observer (NPW)** NPW observer with eye filter (NPWE)*** Incorporates internal noise (referred to the input). Weighs mid-frequencies; Removes DC component of the NEQ. *ICRU Report #5 **Wagner et al. Med. Phys. () 977 *ICRU Report #5 **Wagner et al. Med. Phys. () 977 ***Burgess et al., JOSA. A. () 997 Human observer study Comparison with model observer performance Varied size: 3-5 mm diameter nodules 5 mm nodule 3 mm nodule Varied contrast: -5 HU inserted into the cases experienced radiologists: rated conspicuity from to. *Xiang Li PhD Dissertation (9) % dose Lesion contrast Lesion contrast + d' AUC NPWE: Non-prewhitening (with eye) model observer best matched human observer performance

5 AUC (NPWE) % CNR (water-acrylic) CNR (water-acrylic) Example in CT optimization Example in CT optimization LDPE bone background acrylic kvp D? ma ma >? ma kvp D? ma Limited applicability for IR images Object and dose dependent Observer model results... reconstructions PE... Acrylic -5 ma -5 ma -75 ma - ma - ma... Bone Large feature 7% kvp ACR Acrylic insert. (a).... Spatial Frequency (cycles/mm). (b).... Spatial Frequency (cycles/mm). (c).... Spatial Frequency (cycles/mm) CTDI (mgy). mgy 3.7 mgy 5

6 AUC (NPWE) 9% Observer model results Breast Tomosynthesis Small feature.95 kvp ACR.5 lp/mm bar pattern % CTDI (mgy). mgy 3.7 mgy Breast Tomosynthesis Mammogram Signal performance: 3D measurement: Point source phantom Ensemble averaged across location PSF Coronal Slice Axial Slice Courtesy Joseph Lo and Jay Baker Duke University Spatial frequency (mm - )

7 Detectability index (d ) Detectability index (d ) Noise performance: NPS Detection performance 3D NPS measurement: xx realizations (N=5) Anatomical, quantum, and electronic noise D sphere in coronal slice mm diameter sphere Coronal Slice.5 mgy Axial Slice 5 5 Acquisition angle (degrees) Spatial frequency (mm - ) Detection performance Sphere in coronal slice Variability of Task-based performance mm mass.5 mm calc (%) mm diameter sphere.5 mgy.5 mgy.3 mgy 5 5 Acquisition angle (degrees) Chest wall Overall variability: % Chest Chest wall wall Overall variability:5% 7

8 W task (normalized) /MSE (mm -3 ) / MSE (mm - ) /MSE (mm -3 ) /MSE (mm -3 ) Quantitative Imaging Paradigm shift in imaging: Qualitative Quantitative interpretation of images Clinical Imaging Task* Generating size estimation task function Precision Classification Is a mass present? Is there calcification? Is the lesion benign or malignant? Metrics: SNR ROC (A Z) d' Estimation What size is the lesion? What is the calcium content? What is the density? Metrics: Accuracy (bias) Precision (MSE).5 mgy *ICRU report 5: Assessment of image quality Generating size estimation task function Breast imaging: Area estimation task Precision.5 mgy Precision for Area estimation Precision for Volume estimation.5 mgy Estimation.5 mgy d e.5 mgy.5 mgy d e.3 mgy.3 mgy Spatial frequency (mm - )

9 / MSE (mm - ) /MSE (mm -3 ) Angle (degrees) Breast imaging: Area estimation task Breast imaging: Task dependence Precision for Area estimation Precision for Volume estimation Optimal acquisition angle.5 mgy.5 mgy d e.5 mgy.5 mgy d e.3 mgy.3 mgy Mean glandular dose (mgy) Volume estimation in lung CT Conclusions System performance assessment In scanner measurement of and NPS provided vast amount of information about the imaging system Can be combined to yield a task-based metric for system assessment. Towards clinical implementation Further work required to develop more practical methods of measurements (new phantoms and technique) Need to implement standardized evaluation and interpretation of the metrics (detectability index) 9

10 Acknowledgements RAI LABS