Abstract
The homodyned K-distribution and the K-distribution, viewed as a special case, as well as the Rayleigh and the Rice distributions, viewed as limiting cases, are discussed in the context of quantitative ultrasound (QUS) imaging. The Nakagami distribution is presented as an approximation of the homodyned K-distribution. The main assumptions made are: (1) the absence of log-compression or application of non-linear filtering on the echo envelope of the radiofrequency signal; (2) the randomness and independence of the diffuse scatterers. We explain why other available models are less amenable to a physical interpretation of their parameters. We also present the main methods for the estimation of the statistical parameters of these distributions. We explain why we advocate the methods based on the X-statistics for the Rice and the Nakagami distributions, and the K-distribution. The limitations of the proposed models are presented. Several new results are included in the discussion sections, with proofs in the appendix.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Abraham DA, Lyons AP (2002) Novel physical interpretation of K-distributed reverberation. IEEE J Oceanic Eng 27(4):800–813
Abraham DA, Lyons AP (2010) Reliable methods for estimating the K-distribution shape parameter. IEEE J Oceanic Eng 35(2):288–302
Abramowitz M, Stegun IA (eds) (1972) Handbook of mathematical functions with formulas, graphs, and mathematical tables. Dover, New York
Agrawal R, Karmeshu (2006) Ultrasonic basckscattering in tissue: characterization through Nakagami-generalized inverse Gaussian distribution. Comp Biol Med 37:166–172
Barakat R (1986) Weak-scatterer generalizations of the K-density function with application to laser in atmospheric turbulence. J Opt Soc Am A 3:401–409
Berger NE, Lucas RJ, Twersky V (1991) Polydisperse scattering theory and comparisons with data for red blood cells. J Acoust Soc Am 89(3):1394–1401
Blacknell D, Tough RJA (2001) Parameter estimation for the k-distributed based on [z log (z)]. IEE Proc Radar Sonar Navig 148(6):309–312
Bouhlel N, Sevestre-Ghalila S (2009) Nakagami Markov random field as texture model for ultrasound RF envelope image. Computers Biol Med 39:535–544
Burckhardt CB (1978) Speckle in ultrasound b-mode scans. IEEE Trans Son Ultrason SU-25(1):1–6
Carrobi CFM, Cati M (2008) The absolute maximum of the likelihood function of the Rice distribution: existence and uniqueness. IEEE Trans Instrum Measur 57(4):682–689
Chung P-J, Roberts WJJ, Bohme JF (2005) Recursive K-distribution parameter estimation. IEEE Trans Signal Processing 53(2):397–402
Destrempes F, Meunier J, Giroux M-F, Soulez G, Cloutier G (2009) Segmentation in ultrasonic B-mode images of healthy carotid arteries using mixtures of Nakagami distributions and stochastic optimization. IEEE Trans Med Imag 28(2):215–229
Destrempes F, Cloutier G (2010) A critical review and uniformized representation of statistical distributions modeling the ultrasound echo envelope. Ultrasound Med Biol 36(7):1037–1051
Destrempes F, Meunier J, Giroux M-F, Soulez G, Cloutier G (2011) Segmentation of plaques in sequences of ultrasonic B-mode images of carotid arteries based on motion estimation and a Bayesian model. IEEE Trans Biomed Eng 58(8):2202–2211
Dumane VA, Shankar PM (2001) Use of frequency diversity and Nakagami statistics in ultrasonic tissue characterization. IEEE Trans Ultrason Ferroelectr Freq Control 48(5):1139–1146
Dutt V (1995) Statistical analysis of ultrasound echo envelope. PhD Thesis, Mayo Graduate School, August
Dutt V, Greenleaf JF (1994) Ultrasound echo envelope analysis using a homodyned K distribution signal model. Ultrasonic Imaging 16:265–287
Dutt V, Greenleaf JF (1995) Speckle analysis using signal to noise ratios based on fractional order moments. Ultrasonic Imaging 17:251–268
Dutt V, Greenleaf JF (1996) Statistics of the log-compressed echo envelope. J Acoust Soc Am 99(6):265–287
Edgeworth FY (1908) On the probable errors of frequency-constants I, II, III. J Roy Statist Soc 71:381–397, 499–512, 651–678
Edgeworth FY (1909) On the probable errors of frequency-constants. Addendum. J Roy Statist Soc 72:81–90
Eltoft T (2005) The rician inverse Gaussian distribution: a new model for non-rayleigh signal amplitude statistics. IEEE Trans Image Proces 14:1722–1735
Erdélyi A (1954) Tables of integral transforms. McGraw-Hill Book Co., Inc., New York
Fisher RA (1912) On an absolute criterion for fitting frequency curves. Messenger Math 41:155–160
Fisher RA (1922) On the mathematical foundations of theoretical statistics. Philos Trans Roy London Ser A 222:309–368
Fisher RA (1925) Theory of statistical estimation. In: Proceedings of the cambridge philosophical society, vol 22, pp 700–725
Fisher RA (1956) Statistical methods and scientific inference. Oliver and Boyd, Edinburgh
Gradshteyn IS, Ryshik IM (eds) (1994) Table of integrals, series, and products. Academic Press
Hao X, Bruce CJ, Pislaru C, Greenleaf JF (2002) Characterization of reperfused infarcted myocardium from high-frequency intracardiac ultrasound imaging using homodyned K distribution. IEEE Trans Ultrason Ferroelectr Freq Control 49(11):1530–1542
Hayley SW, Kays TH, Twersky V (1967) Comparison of distribution functions from scattering data on different sets of spheres. IEEE Trans Antennas Propag AP-15(1):118–135
Hruska DP (2009) Improved techniques for statistical analysis of the envelope of backscattered ultrasound using the homodyned K distribution. M.Sc. Thesis, University of Illinois at Urbana-Champaign
Hruska DP, Sanchez J, Oelze ML (2009) Improved diagnostics through quantitative ultrasound imaging. In: International Conference of the IEEE Engineering in Medicine and Biology Society, pp 1956–1959
Hruska DP, Oelze ML (2009) Improved parameter estimates based on the homodyned K distribution. IEEE Trans Ultrason Ferroelectr Freq Control 56(11):2471–2481
Huang C-C, Tsui P-H, Wang S-H (2007) Detection of coagulating blood under steady flow by statistical analysis of backscattered signals. IEEE Trans Ultrason Ferroelectr Freq Control 54(2):435–442
Huang C-C, Wang S-H (2007) Statistical variations of ultrasound signals backscattered from flowing blood. Ultrasound Med Biol 33(12):1943–1954
Insana MF, Wagner RF, Garra BS, Brown DG, Shawker TH (1986) Analysis of ultrasound image texture via generalized Rician statistics. Opt Eng 25(6):743–748
Iskander DR, Zoubir AM (1999) Estimation of the parameters of the K-distribution using higher order and fractional moments. IEEE Trans Aerosp Electron Syst 35(4):1453–1457
Jakeman E, Pusey PN (1976) A model for non-Rayleigh sea echo. IEEE Trans Antennas Propag 24:806–814
Jakeman E (1980) On the statistics of K-distributed noise. J Phys A 13:31–48
Jakeman E, Tough RJA (1987) Generalized k distribution: a statistical model for weak scattering. J Opt Soc Am A 4:1764–1772
Jeffreys H (1946) An invariant form for the prior probability in estimation problems. In: Proc. Roy. Soc. London (Ser. A), vol 186, pp 453–461
Jensen J (1906) Sur les fonctions convexes et les inégalités entre les valeurs moyennes. Acta Mathematica 30:175–193
Joughin IR, Percival DB, Winebrenner DP (1993) Maximum likelihood estimation of K distribution parameters for SAR data. IEEE Trans Geosci Remote Sens 31(5):989–999
Keyes TK, Tucker WT (1999) The K-distribution for modeling the envelope amplitude of a backscattered signal. IEEE Trans Ultrason Ferroelectr Freq Control 46(4):883–887
Kullback S, Leibler RA (1951) On information and sufficiency. Ann Math Stat 22:79–86
Lord RD (1954) The use of the Hankel transform in statistics. I. Biometrika 41:44–55
Lucas RJ, Twersky V (1987) Inversion of ultrasonic scattering data for red blood cell suspension under different flow conditions. J Acoust Soc Am 82(3):794–799
Luke YL (1962) Integrals of Bessel functions. McGraw-Hill Book Company, Inc.
Mamou J, Coron A, Oelze ML, Saegusa-Beecroft E, Hata M, Lee P, Machi J, Yanagihara E, Laugier P, Feleppa EJ (2010) Three-dimensional high-frequency spectral and envelope quantification of excised human lymph nodes. In: IEEE ultrasonics, symposium, pp 604–607
Mamou J, Coron A, Oelze ML, Saegusa-Beecroft E, Hata M, Lee P, Machi J, Yanagihara E, Laugier P, Feleppa EJ (2011) Three-dimensional high-frequency backscatter and envelope quantification of cancerous human lymph nodes. Ultrasound Med Biol 37(3):2055–2068
Martin-Fernandez M, Alberola-Lopez C (2007) Parameter estimation of the homodyned K distribution based on signal to noise ratio. In: IEEE utrasound, symposium, pp 158–161
Molthen RC, Narayanan VM, Shankar PM, Reid JM, Genis V, Vergara-Dominguez L (1993) Ultrasound echo evaluation by K-distribution. In: IEEE ultrasonics, symposium, pp. 957–960
Molthen RC, Shankar PM, Reid JM (1995) Characterization of ultrasonic B-scans using non-Rayleigh statistics. Ultrasound Med Biol 21:161–170
Nakagami M (1940) Study of the resultant amplitude of many vibrations whose phases and amplitudes are at random. J Inst Elec Commun Eng Japan 24(202):17–26
Nakagami M (1943) Statistical character of short-wave fading. J Inst Elec Commun Eng Japan 27:145–150
Nakagami M (1960) The m distribution—a general formula of intensity distribution in rapid fading. In: Hoffman WC (ed) Stat Methods Radio Wave Propag. Pergamon Press, New York, pp 3–36
Narayanan VM, Shankar PM, Reid JM (1994) Non-Rayleigh statistics of ultrasonic backscattered signals. IEEE Trans Ultrason Ferroelectr Freq Control 41(6):845–852
Narayanan VM, Molthen RC, Shankar PM, Vergara L, Reid JM (1997) Studies on ultrasonic scattering from quasi-periodic structures. IEEE Trans Ultrason Ferroelectr Freq Control 44(1):114–124
Nillesen MM, Lopata RGP, Gerrits IH, Kapusta L, Thussen JM, de Korte CL (2008) Modeling envelope statistics of blood and myocardium for segmentation of echocardiographic images. Ultrasound Med Biol 34(4):674–680
Oelze ML, O’Brien WD Jr (2007) Quantitative ultrasound assessment of breast cancer using a multiparameter approach. In: IEEE utrasound symposium, pp 981–984
Oliver CJ (1993) Optimum texture estimators for SAR clutter. J Phys D Appl Phys 26(11):1824–1835
Prager RW, Gee AH, Treece GM, Berman LH (2002) Analysis of speckle in ultrasound images using fractional order statistics and the homodyned K-distribution. Ultrasonics 40:133–137
Prager RW, Gee AH, Treece GM, Berman LH (2003) Decompression and speckle detection for ultrasound images using the homodyned K-distribution. Pattern Recogn Let 24(4–5):705–713
Pratt JW (1976) FY Edgeworth and RA Fisher on the efficiency of maximum likelihood estimation. Ann Stat 4:501–514
Raju BI, Srinivasan MA (2002) Statistics of envelope of high-frequency ultrasonic backscatter from human skin in vivo. IEEE Trans Ultrason Ferroelect Freq Contr 49(7):871–882
Rayleigh L (1880) On the resultant of a large number of vibrations of the same pitch and of arbitrary phase. Phil Mag 10:73
Redner R (1981) Note on the consistency of the maximum likelihood estimate for nonidentifiable distributions. Ann Stat 9(1):225–228
Redner RA, Walker HF (1984) Mixture densities, maximum likelihood and the EM algorithm. SIAM Rev 26(2):195–239
Rice SO (1945) Mathematical analysis of random noise. Bell Sys Tech J 24:46–156
Rice SO (1954) Mathematical analysis of random noise. In: Wax N (ed) Selected papers on noise and stochastic processes. Dover, New York, pp 133–294
Roberts WJJ, Furui S (2000) Maximum likelihood estimation of K-distribution parameters via the expectation-maximization algorithm. IEEE Trans Signal Proces 48(12):3303–3306
Saha RK, Kolios MC (2011) Effects of cell spatial organization and size distribution on ultrasound backscattering. IEEE Trans Ultrason Ferroelectr Freq Control 58(10):2118–2131
Shankar PM, Reid JM, Ortega H, Piccoli CW, Goldberg BB (1993) Use of non-Rayleigh statistics for the identification of tumors in ultrasonic B-scans of the breast. IEEE Trans Med Imag 12(4):687–692
Shankar PM (1995) A model for ultrasonic scattering from tissues based on the K distribution. Phys Med Biol 40:1633–1649
Shankar PM (2000) A general statistical model for ultrasonic backscattering from tissues. IEEE Trans Ultrason Ferroelectr Freq Control 47(3):727–736
Shankar PM, Dumane VA, Reid JM, Genis V, Forsberg F, Piccoli CW, Goldberg BB (2001) Classification of ultrasonic B-mode images of breast masses using Nakagami distribution. IEEE Trans Ultrason Ferroelectr Freq Control 48(2):569–580
Shankar PM, Forsberg F, Lown L (2003) Statistical modeling of atherosclerotic plaque in carotid B mode images—a feasibility study. Ultrasound Med Biol 29(9):1305–1309
Shankar PM (2003) A compound scattering pdf for the ultrasonic echo envelope and its relationship to K and Nakagami distributions. IEEE Trans Ultrason Ferroelect Freq Contr 50(3):339–343
Talukdar KK, Lawing WD (1991) Estimation of the parameters of the Rice distribution. J Acoust Soc Am 89(3):1193–1197
Tao Z, Beaty J, Jaffe CC, Tagare HD (2002) Gray level models for segmenting myocaridum and blood in cardiac ultrasound images. In: IEEE international symposium on biomedical, imaging, pp 265–268
Tao Z, Tagare HD, Beaty JD (2006) Evaluation of four probability distribution models for speckle in clinical cardiac ultrasound images. IEEE Trans Med Imag 25(11):1483–1491
Tsui P-H, Wang S-H (2004) The effect of transducer characteristics on the estimation of nakagami paramater as a function of scatterer concentration. Ultrasound Med Biol 30(10):1345–1353
Tsui P-H, Wang S-H, Huang C-C (2005) The effect of logarithmic compression on estimation of the nakagami parameter for ultrasonic tissue characterization: a simulation study. Phys Med Biol 50(14):3235–3244
Tsui P-H, Huang C-C, Chang C-C, Wang S-H, Shung K-K (2007) Feasibility study of using high-frequency ultrasonic Nakagami imaging for characterizing the cataract lens in vitro. Phys Med Biol 52(21):6413–6425
Tsui P-H, Chang C-C (2007) Imaging local scatterer concentrations by the nakagami statistical model. Ultrasound Med Biol 33(4):608–619
Tsui P-H, Yeh C-K, Chang C-C (2008a) Feasibility exploration of blood flow estimation by contrast-assisted Nakagami imaging. Ultrason Imaging 30(3):133–150
Tsui P-H, Yeh C-K, Chang C-C, Liao Y-Y (2008b) Classification of breast masses by ultrasonic Nakagami imaging: a feasibility study. Phys Med Biol 53:6027–6044
Tsui P-H, Yeh C-K, Chang C-C (2009a) Microvascular flow estimation by contrast-assisted ultrasound B-scan and statistical parametric images. IEEE Trans Inf Technol Biomed 13(3):360–369
Tsui P-H, Yeh C-K, Chang C-C (2009b) Microvascular flow estimation by microbubble-assisted nakagami imaging. Ultrasound Med Biol 35(4):653–671
Tsui P-H, Chang C-C, Ho M-C, Lee Y-H, Chen Y-S, Chang C-C, Huang N-E, Wu Z-H, Chang K-J (2009c) Use of Nakagami statistics and empirical mode decomposition for ultrasound tissue characterization by a nonfocused transducer. Ultrasound Med Biol 35(12):2055–2068
Tsui P-H, Hsu C-W, Ho M-C, Chen Y-S, Lin J-J, Chang C-C, Chu C-C (2010a) Three-dimensional ultrasonic Nakagami imaging for tissue characterization. Phys Med Biol 55(19):5849–5866
Tsui P-H, Liao Y-Y, Chang C-C, Kuo W-H, Chang K-J, Yeh C-K (2010b) Classification of benign and malignant breast tumors by 2-d analysis based on contour description and scatterer characterization. IEEE Trans Med Imag 29(2):513–522
Tsui P-H, Yeh C-K, Liao Y-Y, Chang C-C, Kuo W-H, Chang K-J, Chen C-N (2010c) Ultrasonic Nakagami imaging: a strategy to visualize the scatterer properties of benign and malignant breast tumors. Ultrasound Med Biol 36(2):209–217
Twersky V (1975) Transparency of pair-correlated random distributions of small scatterers with applications to the cornea. J Opt Soc Am 65(5):524–530
Twersky V (1978) Acoustic bulk parameters in distributions of pair-correlated scatterers. J Opt Soc Am 64:1710–1719
Twersky V (1987) Low-frequency scattering by correlated distributions of randomly oriented particles. J Acoust Soc Am 81(7):1609–1618
Twersky V (1988) Low-frequency scattering by mixtures of correlated nonspherical particles. J Acoust Soc Am 84(1):409–415
Wagner RF, Smith SW, Sandrick JM, Lopez H (1983) Statistics of speckle in ultrasound B-scans. IEEE Trans Sonics Ultras 30(3):156–163
Wagner RF, Insana MF, Brown DG (1987) Statistical properties of radio-frequency and envelope-detected signals with applications to medical ultrasound. J Opt Soc Am A 4(5):910–922
Weng L, Reid J, Shankar P, Soetanto K (1990) Non-uniform ultrasound speckle phase distribution applied to scatterer spacing estimation. In: IEEE ultrasonics, symposium, pp 1593–1596
Weng L, Reid JM, Shankar PM, Soetanto K, Lu X-M (1992) Nonuniform phase distribution in ultrasound speckle analysis-Part I: background and experimental demonstration. IEEE Trans Ultrason Ferroelectr Freq Control 39(3):352–359
Yamaguchi T, Zenbutsu S, Igarashi Y, Kamiyama N, Mamou J, Hachiya H (2011) Echo envelope analysis method for quantifying heterogeneity of scatterer distribution for tissue characterization of liver fibrosis. In: IEEE ultrasonics, symposium, pp 1412–1415
Acknowledgments
The authors acknowledge the continuous financial support of the Canadian Institutes of Health Research and Natural Sciences and Engineering Research Council of Canada.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Appendix: Proofs of the New Results
Appendix: Proofs of the New Results
Proof of Theorem 8 From Theorem 6, there is exactly one critical point of \(L(\varepsilon ,\sigma ^2)\) for which \(\varepsilon >0\), and it is the MLE (here, for \(L\) denotes \(L_{\text{ Ri }}\)). Therefore (using Theorem 5), the function \(f(\kappa )\) has exactly one positive root \(\kappa _*\) and it corresponds to the MLE. Moreover, one can check that \(\kappa =0\) is also a root of the function \(f\). Namely, we have \(\lim \nolimits_{\kappa \rightarrow 0}f(\kappa )=-1+\frac{1}{N}\sum \nolimits_{i=1}^N y_i^2\), and by construction, \(\frac{1}{N}\sum \nolimits_{i=1}^N y_i^2=1\).
We have \(\lim \nolimits_{\kappa \rightarrow \infty }f(\kappa )=2(-1+\frac{1}{N}\sum\nolimits_{i=1}^N y_i)=2(-1+\overline{\sqrt{I}}/\sqrt{\overline{I}})\). A direct application of Cauchy-Schwartz’ inequality ensures that \(\overline{\sqrt{I}}/\sqrt{\overline{I}}<1\), so that \(\lim \nolimits_{\kappa \rightarrow \infty }f(\kappa )<0\). In view of the Intermediate Value Theorem for continuous functions, it follows that \(f(\kappa )<0\), if \(\kappa >\kappa _*\).
Next, we want to show that \(f(\kappa )>0\) for \(\kappa \in (0,\kappa _*)\). Since \(\kappa _*\) is the only positive root of \(f\), and since \(f(\kappa )<0\) on \((\kappa _*,\infty )\), it is enough to show that \(\frac{\partial f}{\partial \kappa }<0\) at \(\kappa _*\); for then, \(f(\kappa )>0\) if \(\kappa <\kappa _*\) is sufficiently near \(\kappa _*\), and hence, \(f(\kappa )>0\) on \((0,\kappa _*)\) using the Intermediate Value Theorem.
First of all, we claim that \(\frac{\partial f}{\partial \kappa }=\frac{1}{N}\frac{\partial ^2 L}{\partial \kappa ^2}\) at a critical point of \(L(\varepsilon ,\sigma ^2)=\sum\nolimits_{i=1}^N \log P_{\text{ Ri }}(A_i\,|\,\varepsilon ,\sigma ^2)\), whenever \(\varepsilon >0\) (i.e. \(\kappa >0\)). Indeed, with the change of variable \(\varepsilon = \sqrt{\frac{\mu \kappa }{(\kappa +1)}}\) and \(\sigma ^2=\frac{\mu }{2(\kappa +1)}\), we obtain directly from Eq. (10.2)
Next, the derivative of \(\frac{1}{N}L(\mu ,\kappa )\) with respect to \(\kappa \) is equal to
But, from Talukdar and Lawing (1991), we have \(\mu =\overline{I}=\frac{1}{N}\sum\nolimits_{i=1}^N A_i^2\) at a critical point \((\varepsilon ,a^2)\) of \(L_{\text{ Ri }}\). Therefore, we obtain that \(\frac{1}{N}\frac{\partial L}{\partial \kappa }=f(\kappa )\) at such a critical point (because \(A_i/\sqrt{\mu }\) is then equal to \(y_i=A_i/\sqrt{\overline{I}}\)). Taking the partial derivative of Eq. (10.63) with respect to \(\kappa \), we also see that \(\frac{\partial f}{\partial \kappa }=\frac{1}{N}\frac{\partial ^2 L}{\partial \kappa ^2}\) at a critical point \((\varepsilon ,\sigma ^2)\) of \(L\).
Now, recall that if \(u=u(x,y)\) and \(v=v(x,y)\) is a change of variable, then
At this point, we find convenient to use the change of variable \(\varepsilon ^2=\mu \kappa /(1+\kappa )\) and \(\sigma ^2=\mu /(2(1+\kappa ))\). We develop \(\frac{\partial ^2}{\partial \kappa ^2}L=\Bigl ( G_{11} -{G_{12}}+\frac{1}{4}G_{22}\Bigr )\frac{\mu ^2}{(\kappa +1)^4}\) at a critical point of \(L\), where \(G_{11}=\frac{\partial ^2 L}{\partial \varepsilon ^2 \partial \varepsilon ^2}\), \(G_{12}=\frac{\partial ^2 L}{\partial \varepsilon ^2\partial \sigma ^2}\), and \(G_{22}=\frac{\partial ^2 L}{\partial \sigma ^2\partial \sigma ^2}\) (we make use of the fact that \(\frac{\partial L}{\partial \varepsilon ^2}=0=\frac{\partial L}{\partial \sigma ^2}\) at the critical point). Now, from Carrobi and Cati (2008, Appendix A, p. 686-687), we have \(H_{11}H_{22} - H_{12}^2>0\) and \(H_{11}<0\) at the critical point of interest, where \(H_{11}=\frac{\partial ^2 L}{\partial \varepsilon \partial \varepsilon }\), \(H_{12}=\frac{\partial ^2 L}{\partial \varepsilon \partial \sigma ^2}\), and \(H_{22}=\frac{\partial ^2 L}{\partial \sigma ^2\partial \sigma ^2}\) (\(\sigma ^2\) is viewed as a variable). From there, if one uses the change of variable \(\varepsilon =\sqrt{\varepsilon ^2}\) (and \(\sigma ^2=\sigma ^2\)), one concludes that \(G_{11}G_{22} -G_{12}^2=\Bigl ( H_{11}H_{22} - H_{12}^2\Bigr )\frac{1}{4\varepsilon ^2}>0\) and \(G_{11}=H_{11}\frac{1}{4\varepsilon ^2}<0\), at that critical point. Thus, we obtain the upper bound \(G_{22}<G_{12}^2/G_{11}\) (because \(G_{11}<0)\), and therefore \(\frac{\partial ^2 L}{\partial \kappa ^2}<\Bigl ( G_{11}-G_{12}+\frac{1}{4}G_{12}^2/G_{11}\Bigr )\frac{\mu ^2}{(\kappa +1)^4}\). But this is equal to \(\frac{1}{G_{11}}\frac{\mu ^2}{(\kappa +1)^4} \Bigl ( G_{11}-\frac{1}{2}G_{12} \Bigr )^2\), and it is non-positive since \(G_{11}<0\). Therefore, \(\frac{\partial f}{\partial \kappa }=\frac{1}{N}\frac{\partial ^2 L}{\partial \kappa ^2}<0\) at the point \(\kappa =\kappa _*\) (with \(\mu =\overline{I}\)). This completes the proof of Theorem 8. \(\blacksquare \)
Proof of Theorem 9 (a) Setting \(\kappa =0\) in Theorem 7, we obtain directly \(M_{\text{ Ri }}^{(\nu )}(0)=\varGamma (\nu /2+1)\).
(b) From Luke (1962, pp.7–8), we have the following asymptotic behavior
where \(\chi =a_1-b_1\), valid for \(|\arg z|<\pi \) and \(|z|\rightarrow \infty \). Therefore, we have
We conclude that \(\lim \nolimits_{\kappa \rightarrow \infty } M_{\text{ Ri }}^{(\nu )}(\kappa )=\lim \nolimits_{\kappa \rightarrow \infty } \frac{\varGamma (\nu /2+1)e^{-\kappa }}{(\kappa +1)^{\nu /2}} \times \frac{1}{\varGamma (\nu /2+1)} \kappa ^{\nu /2} e^{\kappa }=1\).
(c) From the definition \(M_{\text{ Ri }}^{(\nu )}(\kappa )=\varGamma (\nu /2+1) \frac{\, _1F_1(1+\nu /2,1,\kappa )}{e^{\kappa }(\kappa +1)^{\nu /2}}\), we obtain after algebraic simplifications
Now, from Gradshteyn and Ryshik (1994, 9.213, p.1086) and Gradshteyn and Ryshik (1994, 9.212(3), p.1086), we have \(\frac{d}{d\, \kappa }\, _1F_1(1+\nu /2,1,\kappa )=(1+\nu /2) \, _1F_1(2+\nu /2,2,\kappa )= \frac{\nu }{2}\, _1F_1(1+\nu /2,2,\kappa )+\, _1F_1(1+\nu /2,1,\kappa )\). So, omitting the positive factor \(\varGamma (\nu /2+1)e^{-\kappa }(\kappa +1)^{-\nu /2}\), we obtain
Multiplying by \((\kappa +1)\) and dividing by \(\nu /2\) (both are positive numbers), we obtain
Using (Gradshteyn and Ryshik 1994, 9.212(2), p.1086), we have \(\kappa \, _1F_1(1+\nu /2,2,\kappa )-\, _1F_1(1+\nu /2,1,\kappa )=-\, _1F_1(\nu /2,1,\kappa )\). Therefore, we finally obtain (up to a positive constant)
Now, by definition, the hypergeometric function \(\, _1F_1(a,b,z)\) is equal to \(\sum \nolimits_{n=0}^\infty \frac{(a)_n}{(b)_n}\frac{z^n}{n!}\), where \((a)_n=a(a+1)...(a+n-1)\) is the rising factorial. If \(\nu /2<1\), then \(\frac{(1+\nu /2)_n}{(2)_n}>\frac{(\nu /2)_n}{(1)_n}\) and hence \(\, _1F_1(1+\nu /2,2,\gamma ) - \,_1F_1(\nu /2,1,\gamma )>0\), On the other hand, if \(\nu /2>1\), then \(\frac{(1+\nu /2)_n}{(2)_n}<\frac{(\nu /2)_n}{(1)_n}\) and hence \(\, _1F_1(1+\nu /2,2,\kappa ) - \, _1F_1(\nu /2,1,\kappa )<0\). This completes the proof of the theorem. \(\blacksquare \)
Proof of Theorem 10 (a) First of all, using the change of variable \(I=A^2\), one computes
which is a Laplace transform equal to \(\varGamma (0,\frac{\varepsilon ^2}{2\sigma ^2})+\log \varepsilon ^2\), where \(\varGamma (0,x)\) is the incomplete gamma function \(\int _{x}^\infty \frac{e^{-t}}{t}\, d t\). Then, after subtraction by the term \(\log (\varepsilon ^2+2\sigma ^2)\), one obtains \(\varGamma (0,\frac{\varepsilon ^2}{2\sigma ^2})+\log (\frac{\varepsilon ^2}{\varepsilon ^2+2\sigma ^2})\), which is equal to \(\varGamma (0,\kappa )+\log (\frac{\kappa }{\kappa +1})\) (where \(\kappa =\varepsilon ^2/(2\sigma ^2)\)).
(b) Again, using the change of variable \(I=A^2\), we compute
This Laplace transform is equal to \(4\sigma ^2-2e^{-\frac{\varepsilon ^2}{2\sigma ^2}}\sigma ^2+(\varepsilon ^2+2\sigma ^2)\bigl ( \varGamma (0,\frac{\varepsilon ^2}{2\sigma ^2})+\log \varepsilon ^2 \bigr )\). Dividing by \(\varepsilon ^2+2\sigma ^2\) and subtracting \(E[\log I]=\varGamma (0,\frac{\varepsilon ^2}{2\sigma ^2})+\log \varepsilon ^2\) (from the proof of part a)), one obtains \((4\sigma ^2-2e^{-\frac{\varepsilon ^2}{2\sigma ^2}}\sigma ^2)\div (\varepsilon ^2+2\sigma ^2)\), which is equal to \(\frac{1}{\kappa +1}(2-e^{-\kappa })\), after algebraic simplifications. \(\blacksquare \)
Proof of Theorem 11 (a) From Abramowitz and Stegun (1972, (6.5.15), p. 262), we have \(\varGamma (0,\kappa )=E_1(\kappa )\) (the exponential integral). Moreover, from Abramowitz and Stegun (1972, (5.1.11), p. 229), \(E_1(\kappa )=-\gamma _E-\log \kappa +\sum \nolimits_{n=1}^\infty \frac{(-1)^n\kappa ^n}{nn!}\). We conclude that \(U_{\text{ Ri }}(\kappa )=-\gamma _E-\log (1+\kappa )+\sum \nolimits_{n=1}^\infty \frac{(-1)^n\kappa ^n}{nn!}\). Henceforth, \(\lim \nolimits_{\kappa \rightarrow 0} U_{\text{ Ri }}(\kappa )=-\gamma _E\).
(b) Since \(\varGamma (0,\kappa )=\int _\kappa ^\infty \frac{e^{-t}}{t}\, d t\), it follows that \(\lim \nolimits_{\kappa \rightarrow \infty }\varGamma (0,\kappa )=0\). Moreover, \(\lim \nolimits_{\kappa \rightarrow \infty }\log \frac{\kappa }{\kappa +1}=0\).
(c) We compute \(\frac{d}{d\, \kappa }U_{\text{ Ri }}(\kappa )=-\frac{e^{-\kappa }}{\kappa }+\frac{1}{\kappa }-\frac{1}{\kappa +1}\). This is positive because \(e^\kappa >1+\kappa \), for \(\kappa >0\). \(\blacksquare \)
Proof of Theorem 12 Parts (a) and (b) follow from basic Calculus.
(c) We compute \(\frac{d}{d\, \kappa } X_{\text{ Ri }}(\kappa )=\frac{e^{-\kappa }(\kappa +1)-(2-e^{-\kappa })}{(\kappa +1)^2}\). Ignoring the positive factor \(1/(\kappa +1)^2\), we obtain \(e^{-\kappa }(2+\kappa )-2\). This is negative since \(e^{\kappa }>1+\kappa /2\), for \(\kappa >0\). \(\blacksquare \)
Lemma 5
Let \(\alpha >0\) be fixed. Denote any root of \(\frac{\partial }{\partial \sigma ^2}L(\sigma ^2,\alpha )\) by \(\sigma ^2(\alpha ,\tilde{A})\), where \(\tilde{A}=\{A_1,A_2,...,A_N\}\).
-
(a)
If \(0<\alpha \le 1/2\), then \(\sigma ^2(\alpha ,\tilde{A})\ge \frac{\alpha +\alpha ^2+\sqrt{2\alpha ^3+\alpha ^4}}{\alpha ^2}\bigl ( \overline{A}\bigr )^2\).
-
(b)
If \(1/2 < \alpha \le 3\), then \(\sigma ^2(\alpha ,\tilde{A})\ge \frac{1}{2\alpha ^2}\bigl ( \overline{A}\bigr )^2\).
-
(c)
If \(\alpha >3\), then \(\sigma ^2(\alpha ,\tilde{A})\ge \frac{2\alpha -3+\sqrt{4\alpha -7}}{4(\alpha -2)^2}\bigl ( \overline{A}\bigr )^2\).
-
(d)
If \(0<\alpha \le 1/2\), then \(\sigma ^2(\alpha ,\tilde{A})\le \frac{1}{2\alpha ^2}\bigl (\overline{A}\bigr )^2\).
-
(e)
If \(1/2<\alpha \le 3/2\), then \(\sigma ^2(\alpha ,\tilde{A})\le \frac{1}{2(\alpha /2+1/4)^2} \bigl (\overline{A}\bigr )^2\).
-
(f)
If \(3/2<\alpha \le 3\), then \(\sigma ^2(\alpha ,\tilde{A})\le \frac{1}{2} \bigl (\overline{A}\bigr )^2\).
-
(g)
If \(\alpha >3\), then \(\sigma ^2(\alpha ,\tilde{A})\le \frac{1}{2(\alpha -2)}\overline{A^2}\).
-
(h)
The function \(\frac{\partial }{\partial \sigma ^2}L(\sigma ^2,\alpha )\) is positive at the lower bounds mentioned in parts (a) to (c), whereas it is negative at the upper bounds of parts (d) to (g).
Proof
We compute
Part a). If \(0<\alpha \le 1/2\), then \(K_{\alpha -1}(x)=K_{1-\alpha }(x)<K_{1}(x)\) and \(K_\alpha (x)>K_0(x)\) for any \(x> 0\). Also, the inequality \(K_0(x)/K_1(x)>1-\frac{1}{(x+1)}\) holds for any \(x>0\). Therefore, from Eq. (10.73), we obtain \(\sigma ^2 \frac{\frac{\partial }{\partial \sigma ^2}P_{\text{ HK }}(A\,|\,\sigma ^2,\alpha )}{P_{\text{ HK }}(A\,|\,\sigma ^2,\alpha )} > -\alpha + \frac{1}{2}f(\sqrt{\frac{2}{\sigma ^2}}A)\), where \(f(x)=x(1-\frac{1}{(x+1)})\). Thus, we obtain that
Here, \(L\) denotes \(L_{\text{ K }}\). Now, the function \(f(x)\) is convex. Therefore, from Jensen’s inequality (Jensen 1906), we conclude that
But the right-hand side of Eq. (10.75) is positive if \(\sigma ^2<\frac{\alpha +\alpha ^2+\sqrt{2\alpha ^3+\alpha ^4}}{\alpha ^2}\bigl ( \overline{A}\bigr )^2\). This proves part a).
Part b). If \(\alpha >1/2\), then \(K_{\alpha }(x)>K_{\alpha -1}(x)\) for any \(x> 0\). Therefore, from Eq. (10.73), we obtain \(\sigma ^2 \frac{\frac{\partial }{\partial \sigma ^2}P_{\text{ K }}(A\,|\,\sigma ^2,\alpha )}{P_{\text{ K }}(A\,|\,\sigma ^2,\alpha )} > -\alpha + \frac{1}{\sqrt{2\sigma ^2}}A\). Thus, we conclude that
But the right-hand side of Eq. (10.76) is positive if \(\sigma ^2<\frac{1}{2\alpha ^2}\bigl (\overline{A}\bigr )^2\). This proves part b).
Part c). If \(\alpha >3\), then \(\frac{x}{2}\frac{K_\alpha (x)}{K_{\alpha -1}(x)}>\frac{x}{2}\times (\frac{2(\alpha -1)}{x}+\frac{1}{\frac{2(\alpha -2)}{x}+1})\). Thus, \(\sigma ^2 \frac{\frac{\partial }{\partial \sigma ^2}P_{\text{ K }}(A\,|\,\sigma ^2,\alpha )}{P_{\text{ K }}(A\,|\,\sigma ^2,\alpha )}\) has lower bound \(-1+f(\frac{A}{\sqrt{2\sigma ^2}})\), where \(f(x)=(\frac{(\alpha -2)}{x^2}+\frac{1}{x})^{-1}\). Thus, we conclude that
From Jensen’s inequality, we then obtain
because the function \(f(x)\) above is convex. But the right-hand side of Eq. (10.78) is positive if \(\sigma ^2<\frac{2\alpha -3+\sqrt{4\alpha -7}}{4(\alpha -2)^2}\bigl ( \overline{A}\bigr )^2 \). This proves part c).
Part d). If \(0<\alpha \le 1/2\), then \(K_{\alpha }(x)<K_{\alpha -1}(x)\) for any \(x> 0\). Therefore, from Eq. (10.73), we obtain \(\sigma ^2 \frac{\frac{\partial }{\partial \sigma ^2}P_{\text{ K }}(A\,|\,\sigma ^2,\alpha )}{P_{\text{ K }}(A\,|\,\sigma ^2,\alpha )} < -\alpha + \frac{1}{\sqrt{2\sigma ^2}}A\). This yields the inequality
But the right-hand side of Eq. (10.79) is negative if \(\sigma ^2>\frac{1}{2\alpha ^2}\bigl ( \overline{A}\bigr )^2\). This proves part d).
Part e). If \(1/2<\alpha \le 3/2\), then \(\frac{K_{\alpha }(x)}{K_{\alpha -1}(x)} <1+\frac{(\alpha -1/2)}{x}\) for any \(x>0\). Therefore, we have \(\sigma ^2 \frac{\frac{\partial }{\partial \sigma ^2}P_{\text{ K }}(A\,|\,\sigma ^2,\alpha )}{P_{\text{ K }}(A\,|\,\sigma ^2,\alpha )} < -\frac{\alpha }{2}- \frac{1}{4} +\frac{1}{\sqrt{2\sigma ^2}}A\). It follows that
But the right-hand side of Eq. (10.80) is negative if \(\sigma ^2>\frac{1}{2(\alpha /2+1/4)^2}\bigl ( \overline{A}\bigr )^2\). This proves part e).
Part f). If \(3/2<\alpha \le 3\), then \(\frac{K_{\alpha }(x)}{K_{\alpha -1}(x)}< 1+\frac{2(\alpha -1)}{x}\) for any \(x>0\). Thus, we obtain \(\sigma ^2 \frac{\frac{\partial }{\partial \sigma ^2}P_{\text{ K }}(A\,|\,\sigma ^2,\alpha )}{P_{\text{ K }}(A\,|\,\sigma ^2,\alpha )} < -\alpha + (\alpha -1) +\frac{1}{\sqrt{2\sigma ^2}}A\). From there, we conclude that
But the right-hand side of Eq. (10.81) is negative if \(\sigma ^2>\frac{1}{2} \bigl ( \overline{A} \bigr )^2\). Hence, part f) of the Theorem.
Part g). If \(3<\alpha \), then \(\frac{x}{2}\frac{K_{\alpha }(x)}{K_{\alpha -1}(x)}< (\alpha -1)+\frac{x^2}{4(\alpha -2)}\) for any \(x>0\). Therefore, we obtain
But the right-hand side of Eq. (10.82) is negative if \(\sigma ^2>\frac{1}{2(\alpha -2)} \overline{A^2}\). Hence, part g) of the Theorem.
Finally, part h) follows from the proof of parts a) to g). \(\blacksquare \)
Proof of Theorem 18 From Lemma 5, for any \(\alpha >0\), there exist two values \(0<\sigma _1^2<\sigma _2^2\) for which \(\frac{\partial }{\partial \sigma ^2}L(\sigma _1^2,\alpha )>0\) and \(\frac{\partial }{\partial \sigma ^2}L(\sigma _2^2,\alpha )<0\), where \(L\) denotes \(L_{\text{ K }}\). Thus, by the Intermediate Value Theorem, there exists \(\sigma ^2=\sigma ^2(\alpha ,\tilde{A})\) such that \(\frac{\partial }{\partial \sigma ^2}L(\sigma ^2,\alpha )=0\). \(\blacksquare \)
Proof of Theorem 19 Part a). Let \(0<\alpha <1/2\). In Eq. (10.32), the term \(-\psi (\alpha ) +\log \bigl ( \frac{x}{2} \bigr ) - \frac{\frac{\partial }{\partial \alpha }K_{1-\alpha }(x)}{K_{1-\alpha }(x)}\) is an increasing function of \(x>0\). Also, from Lemma 5 part d), we have \(\sigma ^2(\alpha ,\tilde{A})\le \frac{1}{2\alpha ^2} \bigl ( \overline{A} \bigr )^2\). Therefore, we obtain \(\sqrt{\frac{2}{\sigma ^2}}A_i\ge 2 \alpha \frac{A_i}{\overline{A}}\). It follows that \(LB(\alpha )=-\psi (\alpha ) +\log \bigl ( \alpha \frac{A_i}{\overline{A}} \bigr ) - \frac{\frac{\partial }{\partial \alpha }K_{1-\alpha }\bigl (2 \alpha \frac{A_i}{\overline{A}}\bigr )}{K_{1-\alpha }\bigl (2 \alpha \frac{A_i}{\overline{A}}\bigr )}\) is a lower bound for that term. Now, from Abramowitz and Stegun (1972, Eq. (9.6.45), p. 377), we have \(\frac{\frac{\partial }{\partial \alpha }K_{1-\alpha }(x)}{K_{1-\alpha }(x)}\approx \frac{\frac{\partial }{\partial \alpha }K_{1}(x)}{K_{1}(x)}=\frac{K_0(x)}{x\, K_1(x)}\) as \(\alpha \rightarrow 0\). Moreover, from Abramowitz and Stegun (1972, Eqs. (9.6.8) and (9.6.9), p. 375), we have \(\frac{K_0(x)}{x\, K_1(x)}\sim -\log x\) for small values of \(x>0\). But \(x=2 \alpha \frac{A_i}{\overline{A}}\) has small values for \(\alpha \rightarrow 0\). Thus, we obtain \(\lim \nolimits_{\alpha \rightarrow 0} \alpha LB(\alpha )=\lim \nolimits_{\alpha \rightarrow 0} \alpha \Bigl \{ -\psi (\alpha )+\log \bigl ( \alpha \frac{A_i}{\overline{A}} \bigr )+ \log \bigl ( 2 \alpha \frac{A_i}{\overline{A}} \bigr )\Bigr \}=1\). This proves part a).
Part b). First of all, we observe that there exist constants \(0<C_1<C_2\), such that \(\frac{1}{C_2} \le \liminf \nolimits_{\alpha \rightarrow \infty } \frac{\sigma ^2(\alpha ,\tilde{A})}{1/\alpha }\le \limsup \nolimits_{\alpha \rightarrow \infty } \frac{\sigma ^2(\alpha ,\tilde{A})}{1/\alpha } \le \frac{1}{C_1}\). The first inequality follows from Lemma 5 part c), whereas the third inequality follows from Lemma 5 part g).
Let \(L\) denote \(L_{\text{ K }}\). Since by definition \(\frac{\partial }{\partial \sigma ^2}L(\sigma ^2,\alpha )\vert _{\sigma ^2(\alpha ,\tilde{A})}=0\), we might as well consider the expression \(\alpha \frac{\partial }{\partial \alpha }L(\sigma ^2,\alpha )-\sigma ^2 \frac{\partial }{\partial \sigma ^2}L(\sigma ^2,\alpha )\). From Eqs. (10.32) and (10.33), each term of that expression is equal to
From Abraham and Lyons (2002, Eq. (46)), we have \(\frac{\frac{\partial }{\partial \alpha }K_{\alpha -1}(x)}{K_{\alpha -1}(x)} \sim \psi (\alpha -1)-\log (x/2)+\frac{x^2}{4\alpha ^2}\) for large values of \(\alpha \). Also, from Abraham and Lyons (2002, Eq. (45)), we have \(K_{\alpha }(x) \sim \frac{\varGamma (\alpha )}{2(x/2)^\alpha } \Bigl ( 1- \frac{(x/2)^2}{(\alpha -1)(\alpha -2)}\Bigr )^{\alpha -2}\) for large values of \(\alpha \). Therefore, taking \(\sigma ^2=1/(C\alpha )\), we obtain the asymptotic expression
Finally, Eq. (10.84) tends to \(0\) as \(\alpha \) tends to infinity. This proves part b). \(\blacksquare \)
Proof of Theorem 20 Part a). Let \(0<\alpha <1\). We consider again Eq. (10.83). Using the asymptotic forms (for small values of \(x\) and of \(\alpha \)) \(\frac{\frac{\partial }{\partial \alpha }K_{1-\alpha }(x)}{K_{1-\alpha }(x)}\approx \frac{K_0(x)}{x\, K_1(x)}\), \(\frac{K_0(x)}{x\, K_1(x)}\sim -\log x\) and \((\frac{x}{2})\frac{K_0(x)}{K_1(x)} \sim - \frac{x^2}{2}\log x\), and setting \(x=\sqrt{\frac{2}{\sigma ^2}}A_i\) with \(\sigma ^2=1/(C\alpha )\) and \(C=2/\overline{I}\), we obtain the asymptotic expression
Part a) then follows by taking the limit of Eq. (10.85) as \(\alpha \rightarrow 0\).
Part b). Taking \(\sigma ^2=1/(C\alpha )\), where \(C=2/\overline{I}\), into Eq. (10.84), we obtain the limit \(0\) as \(\rightarrow \infty \). This proves part b). \(\blacksquare \)
Proof of Theorem 21 a) At \(\alpha =0\), we have \(\varGamma (\alpha +\nu /2)=\varGamma (\nu /2)\). Also, \(\varGamma (\alpha )\) has a simple pole with residue \(1\) at \(\alpha =0\). Therefore, \(\frac{\varGamma (\alpha +1/2)}{\alpha ^{\nu /2}\varGamma (\alpha )} \sim \varGamma (\nu /2) \alpha ^{1-\nu /2}\) at \(\alpha \approx 0\), which shows part a).
b) Using Sterling’s formula, we have \(\frac{\varGamma (\alpha +\nu /2)}{\alpha ^{\nu /2}\varGamma (\alpha )} \sim \frac{e^{-\alpha -\nu /2}(\alpha +\nu /2)^{\alpha +\nu /2-1/2}}{\alpha ^{\nu /2}e^{-\alpha }\alpha ^{\alpha -1/2}}\), which is equal to \(e^{-\nu /2} \Bigl ( 1+\frac{\nu /2}{\alpha }\Bigr )^\alpha \Bigl ( 1+\frac{\nu /2}{\alpha }\Bigr )^{\nu /2-1/2} \). Therefore, \(\lim \nolimits_{\alpha \rightarrow \infty } \frac{\varGamma (\alpha +\nu /2)}{\alpha ^{\nu /2}\varGamma (\alpha )}=1\), and we are done.
c) Using the logarithmic derivative, we have \(\frac{d}{d\, \alpha }M_{\text{ K }}^{(\nu )}(\alpha )=M_{\text{ K }}^{(\nu )}(\alpha ) \Bigl ( \psi (\alpha +\nu /2) -\psi (\alpha ) - \frac{\nu }{2\alpha } \Bigr )\). Now, we have \(M_{\text{ K }}^{(\nu )}(\alpha )>0\). Also, \(\frac{1}{\alpha }=\psi (\alpha +1)-\psi (\alpha )\), and hence \(\psi (\alpha +\nu /2) -\psi (\alpha ) - \frac{\nu }{2\alpha }=\psi (\alpha +\nu /2)-\Bigl (\frac{\nu }{2} \psi (\alpha +1)+(1-\frac{\nu }{2})\psi (\alpha )\Bigr )\). Since the function \(\psi \) is convex, we conclude that \(\psi (\alpha +\nu /2)-\Bigl (\frac{\nu }{2} \psi (\alpha +1)+(1-\frac{\nu }{2})\psi (\alpha )\Bigr )>0\), if \(\nu /2<1\), whereas it is negative if \(\nu /2>1\). \(\blacksquare \)
Proof of Theorem 22 a) We consider the function \(f(\alpha )=\frac{\varGamma (\nu +1)\varGamma (\nu +\alpha )\varGamma (\alpha )}{\varGamma ^2(\nu /2+1)\varGamma ^2(\nu /2+\alpha )}\), noting that \(R_{\text{ K }}^{(\nu )}=\bigl (f(\alpha )-1\bigr )^{-1/2}\). Now, as \(\alpha \rightarrow 0\), we have \(\varGamma (\alpha )\rightarrow \infty \), whereas \(\frac{\varGamma (\nu +1)\varGamma (\nu +\alpha )}{\varGamma ^2(\nu /2+1)\varGamma ^2(\nu /2+\alpha )}\rightarrow \frac{\varGamma (\nu +1)\varGamma (\nu )}{\varGamma ^2(\nu /2+1)\varGamma ^2(\nu /2)}>0\). This proves part a).
b) Next, using directly Sterling’s formula for \(\varGamma (\nu +\alpha )\), \(\varGamma (\alpha )\) and \(\varGamma (\nu /2+\alpha )\), one finds that \(\lim \nolimits_{\alpha \rightarrow \infty }f(\alpha )=\frac{\varGamma (\nu +1)}{\varGamma ^2(\nu /2+1)}\), which proves part b).
c) Finally, taking the logarithmic derivative of \(f(\alpha )\) yields \(\frac{d\, f}{d\, \alpha }=f(\alpha )\bigl ( \psi (\nu +\alpha ) + \psi (\alpha ) - 2 \psi (\nu /2+\alpha )\bigr )\). This is negative, since \(f(\alpha )>0\) and \(\psi (\nu /2+\alpha )>\frac{1}{2}(\psi (\nu +\alpha ) + \psi (\alpha ))\) (because \(\psi \) is a convex function). It follows that \(f(\alpha )> \lim \nolimits_{\alpha \rightarrow \infty }f(\alpha )=\frac{\varGamma (\nu +1)}{\varGamma ^2(\nu /2+1)}\). We claim that \(g(\nu )=\frac{\varGamma (\nu +1)}{\varGamma ^2(\nu /2+1)}>1\), for any \(\nu >0\). In fact, the function \(g(\nu )\) is increasing (its derivative is equal to \(g(\nu )(\psi (\nu +1)-\psi (\nu /2+1))\)) and \(g(0)=1\). Therefore, \(f(\alpha )>1\), and it follows that \(\bigl (f(\alpha )-1\bigr )^{-1/2}\) is an increasing function. This completes the proof of part c). \(\blacksquare \)
Proof of Theorem 23 This is an immediate consequence of Theorem 28. \(\blacksquare \)
Proof of Theorem 24 Starting with Eq. (10.44), we compute
Using (Erdélyi 1954, I, p. 146, (29)), this is equal to
This completes the proof of Theorem 25. \(\blacksquare \)
Proof of Theorem 26 a) This follows from Theorem 24, part b), and Theorem 13.
b) From Theorem 24, part a), we know that \(M_{\text{ HK }}^{(\nu )}(\gamma ,\alpha )\) is equal to \(\frac{\varGamma (\nu /2+1)}{(\gamma +\alpha )^{\nu /2}} \int _0^\infty w^{\nu /2}\, _1F_1(-\nu /2,1,-\gamma /w)\mathcal{G}(w\,|\,\alpha ,1)\, d w\). From Luke (1962, pp. 7–8), we have \(\varGamma (\nu /2+1)\, _1F_1(-\nu /2,1,-z)=\varGamma (\nu /2+1)e^{-z}\, _1F_1(1+\nu /2,1,z) =z^{\nu /2}(1+O(1/z))\), for large values of \(z\). Let \(\eta >0\) be a real number (arbitrarily small). Take \(z_0\) sufficiently large so that \((1-\eta ) z^{\nu /2}\le \varGamma (\nu /2+1)\, _1F_1(-\nu /2,1,-z)\le (1+\eta ) z^{\nu /2}\), for any \(z\ge z_0\). Then, if \(\gamma /w\ge z_0\), i.e. \(w\le \gamma /z_0\), we have \((1-\eta ) \gamma ^{\nu /2}\le \varGamma (\nu /2+1)w^{\nu /2}\, _1F_1(-\nu /2,1,-\gamma /w)\le (1+\eta ) \gamma ^{\nu /2}\). Therefore, the integral \(\frac{\varGamma (\nu /2+1)}{(\gamma +\alpha )^{\nu /2}} \int _0^{\gamma /z_0} w^{\nu /2}\, _1F_1(-\nu /2,1,-\gamma /w)\mathcal{G}(w\,|\,\alpha ,1)\, d w\) has lower bound \((1-\eta )\frac{\gamma ^{\nu /2}}{(\gamma +\alpha )^{\nu /2}}Pr(w\le \gamma /z_0)\) and upper bound \((1+\eta )\frac{\gamma ^{\nu /2}}{(\gamma +\alpha )^{\nu /2}}Pr(w\le \gamma /z_0)\). On the other hand, the function \(\, _1F_1(-\nu /2,1,-z)\) equals \(1\) at \(z=0\), and hence there is a real number \(C>0\) such that \(0<\, _1F_1(-\nu /2,1,-\gamma /w)\le C\) for any \(w>\gamma /z_0\). Thus, the integral \(\frac{\varGamma (\nu /2+1)}{(\gamma +\alpha )^{\nu /2}} \int _{\gamma /z_0}^\infty w^{\nu /2}\, _1F_1(-\nu /2,1,-\gamma /w)\mathcal{G}(w\,|\,\alpha ,1)\, d w\) has lower bound \(0\) and upper bound \( \frac{\varGamma (\nu /2+1)}{(\gamma +\alpha )^{\nu /2}} C \frac{\varGamma (\nu /2+\alpha )}{\varGamma (\alpha )}\). But now, \(\lim \nolimits_{\gamma \rightarrow \infty } \frac{\gamma ^{\nu /2}}{(\gamma +\alpha )^{\nu /2}}=1\), \(\lim \nolimits_{\gamma \rightarrow \infty } Pr(w\le \gamma /z_0)=1\), and \(\lim \nolimits_{\gamma \rightarrow \infty } \frac{1}{(\gamma +\alpha )^{\nu /2}}=0\). Therefore, we obtain \(\liminf \nolimits_{\gamma \rightarrow \infty }M_{\text{ HK }}^{(\nu )}(\gamma ,\alpha )\ge 1-\eta \) and \(\limsup \nolimits_{\gamma \rightarrow \infty }M_{\text{ HK }}^{(\nu )}(\gamma ,\alpha )\le 1+\eta \). Since \(\eta \) is arbitrarily small, we conclude that \(\lim \nolimits_{\gamma \rightarrow \infty }M_{\text{ HK }}^{(\nu )}(\gamma ,\alpha )=1\).
c) We consider the function \(f(\gamma ,w)=\frac{\, _1F_1(1+\nu /2,1,\gamma /w)}{e^{\gamma /w}(\gamma +\alpha )^{\nu /2}}\). From Theorem 24, part a), we have \(M_{\text{ HK }}^{(\nu )}(\gamma ,\alpha )=\varGamma (\nu /2+1) \int _0^\infty w^{\nu /2}f(\gamma ,w)\mathcal{G}(w\,|\,\alpha ,1)\, d w\). Thus, we obtain \(\frac{\partial }{\partial \, \gamma } M_{\text{ HK }}^{(\nu )}(\gamma ,\alpha )= \varGamma (\nu /2+1) \int _0^\infty w^{\nu /2}\frac{\partial }{\partial \, \gamma } f(\gamma ,w)\mathcal{G}(w\,|\,\alpha ,1)\, d w\).
We compute the value of \(\frac{\partial }{\partial \, \gamma } \, f(\gamma ,w)\) as
Using (Gradshteyn and Ryshik 1994, 9.213, p. 1086) and (Gradshteyn and Ryshik 1994, 9.212(3), p. 1086), we have \(\frac{d}{d\, z}\, _1F_1(1+\nu /2,1,\gamma /w)=(1+\nu /2) \, _1F_1(2+\nu /2,2,\gamma /w)= \frac{\nu }{2}\, _1F_1(1+\nu /2,2,\gamma /w)+\, _1F_1(1+\nu /2,1,\gamma /w)\). So, we obtain after algebraic simplifications
Using (Gradshteyn and Ryshik (1994), 9.212(2), p. 1086), we have \(\frac{\gamma }{w}\, _1F_1(1+\nu /2,2,\gamma /w)-\, _1F_1(1+\nu /2,1,\gamma /w)=-\, _1F_1(\nu /2,1,\gamma /w)\). Therefore, we finally obtain
Now, let \(\nu /2<1\). Then, we obtain the strict lower bound for \(\frac{\partial }{\partial \, \gamma } \, f(\gamma ,w)\)
Consider the function \(g(\gamma ,w)=\varGamma (\nu /2+1)w^{\nu /2}\frac{\alpha \nu /2}{e^{\gamma /w}(\gamma +\alpha )^{\nu /2+1}w} \, _1F_1(1+\nu /2,2,\gamma /w)\). We have shown that
But, \(\frac{w}{\alpha }\mathcal{G}(w\,|\,\alpha ,1)=\mathcal{G}(w\,|\,\alpha +1,1)\). So, we obtain
Thus, we want to show that \(\int _0^\infty g(\gamma ,w)\mathcal{G}(w\,|\,\alpha ,1)\, d w-\int _0^\infty g(\gamma ,w)\mathcal{G}(w\,|\,\alpha +1,1)\, d w\ge 0\). Ignoring the positive factor \(\varGamma (\nu /2+1)\frac{\alpha \nu /2}{(\gamma +\alpha )^{\nu /2+1}}\), we are thus lead to the function \(h(\gamma ,w)=\frac{w^{\nu /2-1}\, _1F_1(1+\nu /2,2,\gamma /w)}{e^{\gamma /w}}\), and we show that \(\int _0^\infty h(\gamma ,w)\mathcal{G}(w\,|\,\alpha ,1)\, d w-\int _0^\infty h(\gamma ,w)\mathcal{G}(w\,|\,\alpha +1,1)\, d w\ge 0\) as follows. In Lemma 6, we show that \(h(\gamma ,w)\) is decreasing in the variable \(w\), if \(\nu /2<1\). Then, in Lemma 7, we show that for any decreasing positive function \(H(w)\), we have \(\int _0^\infty H(w)\mathcal{G}(w\,|\,\alpha ,1)\, d w-\int _0^\infty H(w)\mathcal{G}(w\,|\,\alpha +1,1)\, d w\ge 0\).
Next, let \(\nu /2>1\). Then, we obtain the strict upper bound for \(\frac{\partial }{\partial \, \gamma } \, f(\gamma ,w)\)
The same argument as above (but with reversed inequalities) leads to
where the function \(g(\gamma ,w)\) is defined as above. So, in this case, we want to show that \(\int _0^\infty h(\gamma ,w)\mathcal{G}(w\,|\,\alpha ,1)\, d w-\int _0^\infty h(\gamma ,w)\mathcal{G}(w\,|\,\alpha +1,1)\, d w\le 0\), where \(h(\gamma ,w)\) is defined as above. But, this is implied by Lemmas 6 and 7 (case \(\nu /2>1\)). This completes the proof of the theorem. \(\blacksquare \)
Lemma 6
a) If \(\nu /2<1\), the function \(h(\gamma ,w)=\frac{w^{\nu /2-1}\, _1F_1(1+\nu /2,2,\gamma /w)}{e^{\gamma /w}}\) is decreasing in the variable \(w\).
b) If \(\nu /2>1\), the function \(h(\gamma ,w)\) is increasing in the variable \(w\).
Proof
Using the change of variable \(x=\gamma /w\), we consider the function \(F(x)=\frac{\, _1F_1(1+\nu /2,2,x)}{e^{x}x^{\nu /2-1}}\). So, we want to show that \(F(x)\) is increasing if \(\nu /2<1\) and \(F(x)\) is decreasing if \(\nu /2>1\) (the function \(x=\gamma /w\) is decreasing in the variable \(w\)).
We compute
Using (Gradshteyn and Ryshik 1994, 9.213, p. 1086) and (Gradshteyn and Ryshik 1994, 9.212(3), p. 1086), we have \(\frac{d}{d\, z}\, _1F_1(1+\nu /2,2,x)=\frac{(1+\nu /2)}{2} \, _1F_1(2+\nu /2,3,x)= \frac{(\nu /2-1)}{2}\, _1F_1(1+\nu /2,3,x)+\frac{2}{2}\, _1F_1(1+\nu /2,2,x)\).
So, we obtain after algebraic simplifications
Using (Gradshteyn and Ryshik 1994, 9.212(2), p. 1086), we finally obtain
The result is now clear. \(\blacksquare \)
Lemma 7
a) Let \(H(w)\) be a decreasing positive function. Then, one has \(\int _0^\infty H(w)\mathcal{G}(w\,|\,\alpha ,1)\, d w-\int _0^\infty H(w)\mathcal{G}(w\,|\,\alpha +1,1)\, d w\ge 0\).
b) Let \(H(w)\) be an increasing positive function. Then, one has
\(\int _0^\infty H(w)\mathcal{G}(w\,|\,\alpha ,1)\, d w-\int _0^\infty H(w)\mathcal{G}(w\,|\,\alpha +1,1)\, d w\le 0\).
Proof
a) Since \(H(w)\) is a positive decreasing function, we can approximate it by functions of the form \(\sum \nolimits_{n=1}^N a_n \, B(b_n,w)\), where \(a_n\ge 0\), \(b_n>0\), and \(B(b,w)\) is equal to \(1\), if \(w\in [0,b]\), and \(B(b,w)=0\), if \(w>b\). Now, \(\int _0^\infty B(b,w)\mathcal{G}(w\,|\,\alpha ,1)\, d w=\int _0^b \mathcal{G}(w\,|\,\alpha ,1)\, d w=1-\frac{\varGamma (\alpha ,b)}{\varGamma (\alpha )}\), where \(\varGamma (\alpha ,b)\) is the incomplete Euler gamma function. But the function \(1-\frac{\varGamma (\alpha ,b)}{\varGamma (\alpha )}\) is decreasing. Therefore, \(\int _0^\infty B(b,w)\mathcal{G}(w\,|\,\alpha ,1)\, d w>\int _0^\infty B(b,w)\mathcal{G}(w\,|\,\alpha +1,1)\, d w\), and we are done.
b) Since \(H(w)\) is a positive increasing function,, we can approximate \(H(w)\) by functions of the form \(\sum\nolimits _{n=1}^N a_n \, (1-B(b_n,w))\), where \(a_n\ge 0\), \(b_n>0\). Now, \(\int _0^\infty (1-B(b,w))\mathcal{G}(w\,|\,\alpha ,1)\, d w=\frac{\varGamma (\alpha ,b)}{\varGamma (\alpha )}\), and we are done. \(\blacksquare \)
Proof of Corollary 2
Proof of Corollary 2. Let \(\nu /2<1\). Since \(0<M<1\), we conclude from Theorem 26, using the Intermediate Value Theorem, that for any \(\alpha >0\) such that \(M_\mathrm{K }(\alpha ) \le M\), there is a unique value of \(\gamma \ge 0\) for which \(M_\mathrm{HK }^{(\nu )}(\gamma ,\alpha )=M\). Thus, if \(M \ge \Gamma (\nu /2+1)\), \(\alpha \) has no restrictions, because \(M_\mathrm{K }^{(\nu )}(\alpha )<\Gamma (\nu /2+1)\) for any \(\alpha >0\) (Theorem 21). On the other hand, if \(M_\mathrm{K }(\alpha ) < M\), let \(\alpha _0\) be the unique solution to the equation \(M_\mathrm{K }^{(\nu )}(\alpha _0)=M\) (Theorem 21). Then, using once more Theorem 21, we obtain that \(M_\mathrm{K }^{(\nu )}(\alpha )< M\) if and only if \(\alpha \le \alpha _0\). Henceforth, if \(M_\mathrm{K }^{(\nu )}(\alpha ) < M\), the domain of the function \(\gamma _M^{(\nu )}(\alpha )\) is the interval \((0,\alpha _0]\)
The case \(\nu /2>1\) is handled similarly, but with reversed inequalities. \(\blacksquare \)
Proof of Lemma 4
Proof of Lemma 4. Part a). From the definition in Eq. (10.6), the distribution \(P_{\text{ HK }}(A\,|\,\varepsilon ,\sigma ^2,\alpha )\) is equal to \(\int _0^\infty P_{\text{ Ri }}(A\,|\,\varepsilon ,\sigma ^2 w) \mathcal{G}(w\,|\,\alpha ,1) d\, w\). Using the identity \(I_0(z)=\frac{1}{\pi }\int _{0}^{\pi }e^{z \cos \theta }d\, \theta \) from Abramowitz and Stegun (1972, Eq. (9.6.16), p. 376) and the definition of the Rice distribution (10.2) , we can express \(P_{\text{ Ri }}(A\,|\,\varepsilon ,\sigma ^2 w)\) in the form \(\frac{1}{\pi }\int _{0}^{\pi } \frac{A}{\sigma ^2w} \exp \Bigl (\frac{\varepsilon }{\sigma ^2w} A \cos \theta \Bigr ) \exp \Bigl ( -\frac{(\varepsilon ^2+A^2)}{2 \sigma ^2w} \Bigr )d\, \theta \). It follows that \(P_{\text{ HK }}(A\,|\,\varepsilon ,\sigma ^2,\alpha )\) can be written as
which yields Eq. (10.56) after evaluation of the inner integral.
Part b). Using Eq. (10.56), the partial derivative of the homodyned K-distribution with respect to \(\varepsilon \) is equal to
Here, we have used the identity \(z\frac{d}{d\, z} K_{\alpha -1}(z)+(\alpha -1)K_{\alpha -1}(z)=-z\, K_{\alpha -2}(z)\) (Abramowitz and Stegun 1972, Eq. (9.6.26), 2nd identity, p. 376) and algebraic simplifications.
Part c). Using Eq. (10.56), the partial derivative of the homodyned K-distribution with respect to \(\sigma ^2\) is equal to
Here, we have used the identity \(\frac{z}{2}\frac{d}{d\, z} K_{\alpha -1}(z)=-\frac{z}{2} K_{\alpha }(z)+\frac{(\alpha -1)}{2}K_{\alpha -1}(z)\) (Abramowitz and Stegun 1972, Eq. (9.6.26), 4th identity, p. 376) and algebraic simplifications.
Part d). Eq. (10.59) follows from part a) upon taking the logarithmic derivative of the integrand in Eq. (10.56). \(\blacksquare \)
Rights and permissions
Copyright information
© 2013 Springer Science+Business Media Dordrecht
About this chapter
Cite this chapter
Destrempes, F., Cloutier, G. (2013). Review of Envelope Statistics Models for Quantitative Ultrasound Imaging and Tissue Characterization. In: Mamou, J., Oelze, M. (eds) Quantitative Ultrasound in Soft Tissues. Springer, Dordrecht. https://doi.org/10.1007/978-94-007-6952-6_10
Download citation
DOI: https://doi.org/10.1007/978-94-007-6952-6_10
Published:
Publisher Name: Springer, Dordrecht
Print ISBN: 978-94-007-6951-9
Online ISBN: 978-94-007-6952-6
eBook Packages: Biomedical and Life SciencesBiomedical and Life Sciences (R0)