1. Towards a Multi-Parametric Quantitative Ultrasonic Tissue Characterization
Roberto J. Lavarello
Laboratorio de Imágenes Médicas, Sección Electricidad y Electrónica
Pontificia Universidad Católica del Perú, San Miguel, Lima 32, Perú
Visión 2015, Universidad San Martín de Porres, Octubre 15, 2015

2. Example of B-mode of breast images from ultrasound
Which one is malignant (if any)?
Constantini et al., JUM, 2006;25:

3. Example of B-mode of breast images from ultrasound
Irregular hypoechoic mass with angular margins and no posterier acoustic features (BI-RADS - 5)
IDC
Hypoechoic mass with circumscribed margins (BI-RADS - 3)
Fibroadenoma

4. Example of B-mode of breast images from ultrasound
Which one is malignant (if any)?
Constantini et al., JUM, 2006;25:

5. Example of B-mode of breast images from ultrasound
Hypoechoic mass with angular margins and no posterier acoustic features (BI-RADS - 4)
Medullary Carcinoma
Irregular hypoechoic mass with spiculated margins and echogenic halo (BI-RADS - 5)
Benign Sclerotic Lesion

6. Backscatter Coefficient
Theory

7. Backscatter coefficient (BSC)
The BSC (derived from the power spectrum of backscattered data) quantifies the frequency-dependent reflectivity of a medium.
QUS allows to obtain micro-structural information from the imaging target. In particular, the parameter under investigation was the effective scatterer diameter or ESD, which represents the typical size of acoustic impedance variations within the region of analysis. ESD estimates are obtained by fitting measured backscattered spectra to parametric scattering models. However, these models are derived with several assumptions and therefore a mismatch between measurements (in blue) and the models (in red) will exist and in turn will degrade the precision of ESD estimates. The precision can be improved by increasing the ROI size, which usually improves the agreement between measurements and scattering models, but at the price of spatial resolution degradation.
Instrument-dependent (transmitted pulse, diffraction) and attenuation effects are compensated in the BSC estimation process.

8. BSC estimation
BSCs can be estimated from several adjacent gated RF data segments using:
|Sm(f)|2 : power spectrum of a gated data line
|S0(f)|2 : power spectrum from a reference plate
A(f) : attenuation compensation function
Gp = (kR2/2F) : focal gain of the transducer
γ : pressure reflection coefficient of the reference plate
F , A0 : transducer focal distance and surface area
L : gate length
Jn(·) : n-th order Bessel function
𝐵𝑆𝐶 𝑓 =2.17∙𝐷 𝐺 𝑝 ∙ 𝛾 2 𝐹 2 𝐴 0 𝐿 ∙ 𝑆 𝑚 𝑓 𝑆 0 𝑓 2 𝐴 𝑓 ,
𝐷 𝐺 𝑝 = 𝑒 −𝑖 𝐺 𝑝 𝐽 0 𝐺 𝑝 +𝑖 𝐽 1 𝐺 𝑝 −1 2 ,

9. BSC-derived parameter estimation
Two spectral parameters (ESD and EAC) were derived.
ESD (in μm) – related to the typical size of scattering structures.
EAC (in mm-3) – proportional to the scatterer number concentration.
The parameter values were obtained by fitting the estimated BSCs to the theoretical spherical Gaussian model:
L : gate length in mm
f: frequency in MHz
q: ratio of aperture radius to data block depth
𝐵𝑆𝐶 𝑓 =2.89∙𝐿∙𝑞∙𝐸𝐴𝐶 𝑓 4 ∙𝐸𝑆𝐷 𝑓∙𝑞∙𝐸𝑆𝐷 𝑒 −3.04 𝑓 2 𝐸𝑆𝐷 2 ,

10. Backscatter Coefficient
Application – Thyroid Cancer
R. Lavarello, B. Ridgway, S. Sarwate, and M. Oelze, “Imaging of follicular variant papillary thyroid carcinoma in a rodent model using spectral-based quantitative ultrasound techniques,” in Proceedings of the IEEE International Symposium on Biomedical Imaging, pp , 2013.
R. Lavarello, B. Ridgway, S. Sarwate and M. Oelze, “Characterization of thyroid cancer in mouse models using high-frequency quantitative ultrasound techniques,” Ultrasound in Medicine and Biology, vol. 39, no. 2, pp , December 2013.

11. Motivation
There is a relatively high prevalence of thyroid nodules in the USA population.
Prevalence estimated to be 2-6% with palpation, 19-25% with ultrasound.
However, thyroid cancer has a low prevalence.
Less than 0.1% prevalence according to the SEER database.
Fine needle aspiration (FNA) biopsy remains the most sensitive and accurate test for thyroid nodule malignancy.
Due to the statistics above, many FNA biopsies result in benign diagnosis.
The goal of this study was to study the feasibility of integrating QUS and ultrasonic tomography imaging by addressing two issues: first, a preliminary implementation of QUS imaging on a UCT scanner was obtained, and second, the use of full angular spatial compounding for extending the QUS spatial resolution vs. precision trade-off was studied.

12. Motivation
Markers derived from conventional ultrasonic methods (echography, Doppler) do not provide sufficient diagnostic accuracy.
Sensitivity, specificity and accuracy around 70-80% when using multiple ultrasound parameters.
Further quantitative information can be derived from ultrasound data.
Parameters derived from backscatter coefficients (BSCs) have shown potential for cancer characterization.
Objective: To explore the usefulness of BSC-derived parameters for characterizing follicular variant papillary thyroid carcinoma (FV-PTC) in a rodent model.
The goal of this study was to study the feasibility of integrating QUS and ultrasonic tomography imaging by addressing two issues: first, a preliminary implementation of QUS imaging on a UCT scanner was obtained, and second, the use of full angular spatial compounding for extending the QUS spatial resolution vs. precision trade-off was studied.

13. Experimental methods
TRβPV/PV mice (Cheng’s Lab, Center for Cancer Research, NIH) were used as a model for FV-PTC.
13 mice (5 diseased, 8 control) were used in this study.
Ultrasound data were obtained from excised thyroids.
In order to avoid effects from intervening tissues.
Data was collected using a 40 MHz, single element transducer.
3 mm diameter, f/3
BSCs were obtained from data blocks of 0.5 by 0.5 mm.
The goal of this study was to study the feasibility of integrating QUS and ultrasonic tomography imaging by addressing two issues: first, a preliminary implementation of QUS imaging on a UCT scanner was obtained, and second, the use of full angular spatial compounding for extending the QUS spatial resolution vs. precision trade-off was studied.

14. Results – QUS images
An alternative approach is to obtain several ESD estimates from the same ROI (in green) with the ultrasonic aperture (in black) at different illumination angles. If the individual estimates are uncorrelated, they can be combined to reduce the variance of ESD images. This idea was originally proposed to explore the limited compounding capabilities of linear arrays, but here it is taken to the limit by collecting data over 360 degrees.
B-mode (left), ESD (center) and EAC (right) images of a normal (top) and diseased (bottom) mice.

15. Results – QUS estimates
Scatter plot of ESD vs. EAC for all animals considered in this study.

16. Thyroid histology images
Follicle size
 Benign : μm
FV-PTC : μm 
Cells
Benign : ≈ 10 μm
FV-PTC : μm 
The simulation results here show the potential of QUS imaging with full angular compounding when imaging inhomogeneous samples using a two-region phantom. On the left, the inclusion is obscured by the large image variance. On the right, using 64 angles of view allowed to improve the precision without degrading the spatial resolution.
Normal thyroid
FV-PTC

17. Thyroid histology images
In practice, full angular spatial compounding can be implemented on an ultrasonic computed tomography scanner. Shown here is an schematic of the scanner used in this study, developed by Techniscan Inc.
Normal thyroid
FV-PTC
Histology images at 40x amplification, showing a higher concentration of cells on the FV-PTC when compared to the normal case.

18. Results – Extended database

19. Discussion
The estimated BSC-derived parameters were consistent with the histological appearance of the studied thyroids.
ESD : the dominant structures were small malignant cells for the FV-PTC cases and large follicles for the normal cases.
EAC : the FV-PTC cases had a larger concentration of scatterers than the normal cases.
Statistically significant differences (p<0.05) were observed between the normal and FV-PTC thyroids.
A non-parametric Kruskal-Wallis test was used due to the small population size.

20. Conclusions
The preliminary experimental results obtained in this work suggest that spectral-based quantitative ultrasonic imaging has the potential for differentiating between normal and malignant thyroid tissues.
Further, these results suggest that quantitative ultrasonic imaging may be sensitive to the changes in tissue architecture resulting from thyroid cancer.

21. Attenuation Coefficient
Theory

22. Attenuation estimation
Attenuation can be estimated using pulse-echo data using the spectral log difference (SLD) algorithm.
SLD estimates the attenuation coefficient from estimates of backscattered power spectra in two blocks (proximal and distal) within a region of interest.

23. Spectral log difference method
The ratio of the power spectra from the distal an proximal blocks can be expressed as
𝑉 𝑠 𝑟 𝑑 ,𝜔 𝑉 𝑠 𝑟 𝑝 ,𝜔 = 𝐴 𝑠 𝑟 𝑑 ,𝜔 ∙ 𝐷 𝑠 𝑟 𝑑 ,𝜔 ∙𝜂( 𝑟 𝑑 ,𝜔) 𝐴 𝑠 𝑟 𝑝 ,𝜔 ∙ 𝐷 𝑠 𝑟 𝑝 ,𝜔 ∙𝜂( 𝑟 𝑝 ,𝜔)
(1)
|𝑉𝑠( 𝑟 ,𝜔)|2 : The power spectrum of the ultrasound data received from a region
centered at depth 𝑟
𝜂( 𝑟 ,𝜔) : The backscatter coefficient.
𝐴 𝑠 𝑟 ,𝜔 : The function describing the total attenuation effects.
𝐷 𝑠 𝑟 ,𝜔 : The mean diffraction correction coefficient

24. Spectral log difference method
By making the assumption that the effective scatterer size is constant so that the BSC varies in the form 𝜂 𝑟 𝑝 ,𝜔 =𝑐∙𝜂( 𝑟 𝑑 ,𝜔), the attenuation coefficient can be found from
Problem: How to evaluate the diffraction term in the equation above?
Reference phantom method
Analytical expression
𝛼 𝜔 + ln 𝑐 4∙ 𝑟 𝑑 − 𝑟 𝑝 = 1 4∙ 𝑟 𝑑 − 𝑟 𝑝 ln 𝑉 𝑠 𝑟 𝑝 ,𝜔 𝑉 𝑠 𝑟 𝑑 ,𝜔 ∙ 𝐷 𝑠 𝑟 𝑑 ,𝜔 𝐷 𝑠 𝑟 𝑝 ,𝜔

25. Analytic diffraction correction function
In this work, the analytic expressions derived by Chen et al. were used for diffraction correction.
𝐷 𝑠 𝑟 ,𝜔 ≅ 0.46∙ 𝜋 𝑎 2 𝑟 2 exp −0.46 𝜋 ∙ 𝐺 𝑝 2 ∙ 𝑓 𝑙 𝑟 − , 1+𝜋 𝐺 𝑝 −1 < 𝑟 𝑓 𝑙 < 1−𝜋 𝐺 𝑝 −1 , 𝜋 𝑎 2 𝑟 2 ∙ 1.07∙ 𝐺 𝑝 ∙ 𝑓 𝑙 𝑟 −1 − , otherwise,
(3)
𝐺 𝑝 : Pressure gain factor.
𝑎: Transducer radius.
𝑓 𝑙 : Focal length.
r : Distance from the transducer to the sample.
X.Chen, D. Phillips, K. Schwarz, J. Mottley and K. Parker “The measurement of backscatter from a broadband pulse-echo system - A new formulation,” IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, vol. 44, pp , 1997

26. Attenuation Coefficient
Application – Thyroid Cancer
O. Zenteno, B. Ridgway, S. Sarwate, M. Oelze, and R. Lavarello, “Ultrasonic attenuation imaging in a rodent thyroid cancer model,” in Proceedings of the IEEE International Ultrasonics Symposium, pp , 2013.
O. Zenteno, A. Luchies, M. Oelze, and R. Lavarello, “Improving the quality of attenuation imaging using full angular spatial compounding,” in Proceedings of the IEEE International Ultrasonics Symposium, pp , 2014.

27. Experimental setup
A weakly focused (f/3) single element transducer with central frequency of 40 MHz and focal length of 9 mm was selected.
Extracted thyroid lobes were placed in a tank of 0.9% saline solution for ultrasound scanning using a micro positioning system.
Several ROIs of 0.6 by 0.6 mm where distributed over the sample with an overlap of 87.5%.
Histopatologic result blind ultrasound image analysis was performed to avoid biasing.

28. Sample attenuation images

29. Distribution of attenuation estimates
Mean/STD: 1.32 ± 0.20 dB/(MHz·cm)
P-values:
C-cell vs. all: p<=
PTC vs. FV-PTC: p =

30. Discussion The estimated mean values are consistent with:
Independent insertion loss measurements obtained from mice (1.19 ± 0.26 dB/MHz·cm).
Reports of attenuation in human thyroids (0.9 dB/MHz·cm to 1.8 dB/MHz·cm).
Attenuation coefficient slope estimates has potential for thyroid characterization.
But high variance may compromise their diagnostic value…

31. Discussion – angular compounding

32. Discussion – angular compounding
Attenuation images from a two-region phantom. Background and inclussion attenuation coefficient slopes were measured to be 0.5 and 1 dB/cm/MHz, respectively. Estimation variance was reduced by 89% using angular compounding.

33. Application – Thyroid Cancer
Towards Human Studies
Application – Thyroid Cancer
J. Rouyer, T. Cueva, T. Yamamoto, A. Portal and R. Lavarello, “In vivo estimation of attenuation and backscatter coefficients from human thyroids,” IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, submitted, 2015.
J. Rouyer, T. Cueva, A. Portal, T. Yamamoto, and R. Lavarello, “Attenuation coefficient estimation of the healthy human thyroid in vivo,” Physics Procedia, vol. 70, pp , 2015.
T. Cueva, J. Rouyer, A. Portal, T. Yamamoto, and R. Lavarello, “Feasibility of quantitative backscatter imaging of human thyroids in vivo,” in Proceedings of the IEEE International Ultrasonics Symposium, accepted for publication, 2015.

34. Data Acquisition and Processing

35. Data Acquisition and Processing

36. Attenuation and Backscatter Coefficients

37. QUS Images in vivo

38. Conclusions
Quantitative ultrasonic attenuation imaging allows some discrimination among four studied thyroid cancer animal models:
c-cell adenoma from normal and malignant groups
Both malignant groups ( i.e., FV-PTC vs. PTC )
Attenuation coefficient slope estimates in conjunction with other techniques (i.e., BSC-based parameters, angular compounding) may have the potential to improve ultrasound-based tissue characterization.

39. Acknowledgements

40. M.Sc. In Digital Signal and Image Processing
Between 2010 and 2014, students from the M.Sc. Program in DSP from PUCP have:
Performed 12 internships in universities and research centers in Europe and North America (USA and Canada)
Published more than 20 articles in indexed, peer-reviewed proceedings of international conferences as first authors
Published three publications in indexed, peer reviewed journals as first authors
Obtained more than 20 awards and scholarships from the PUCP Graduate School, the Peruvian goverment, and the Fulbright program

41. DSP Program students – international activities
9th IEEE International Symposium on Biomedical Imaging, Barcelona, España, May 2012

42. DSP Program students – international activities
IEEE International Ultrasonics Symposium, Dresden, Alemania, October 2012

43. DSP Program students – international activities
12th International Tissue Elasticity Conference, Lingfield, Inglaterra, October 2013

44. DSP Program students – international activities
 11th IEEE EMBS Summer School on Biomedical Imaging, Esmerald Coast, Brittany, June 2014

45. DSP Program students – international activities
IEEE International Ultrasonics Symposium, Chicago, EE.UU., September 2014

46. Thanks!!!