1

Frequency-Radial Duality Based Photoacoustic Reconstruction

S.M. Akramus Salehin, Student Member, IEEE, and Thushara D. Abhayapala, Senior Member, IEEE

AbstractThe abstract goes here.

A Index TermsIEEEtran, journal, L TEX, paper, template.

I. I NTRODUCTION II. BACKGROUND

ON

The pressure recorded by a ultrasound transducer in the time domain and at a position r s can be provided form the general solution to the inhomogeneous Helmholtz equation (1) [3], which is expressed as p(rs , t) = t p0 (r )
V

P HOTOACOUSTIC I MAGING

This section provides a brief overview of photoacoustic imaging together with the wave equations associated with it. Photoacoustic imaging is concerned with imaging medium with an inhomogeneous electromagnetic (EM) absorption and a relatively homogeneous acoustic property. This method of imaging uses the pressure wave generated by the absorption of a laser pulse. It is generally assumed that the incident EM wave is absorbed instantaneously within the target volume of tissue. This condition can be met when the incident laser pulse duration is short such that the temporal illumination function can be approximated by a Dirac delta function (t). Furthermore, a short pulse duration ensures that heat loss and stress dissipation are negligible over the pulse duration. Therefore, assuming constant speed of propagation c in the medium, the pressure p(r , t) at a vector position r and time t in a lossless, linear medium is described by the inhomogeneous Helmholtzs equation [1], [2] p(r, t) c2 2 p(r, t) = p0 (r ) (t) (1) 2t t where the time derivative of the impulsive, initial spatial, pressure distribution p0 (r ) (t) provides the driving force. We will refer to p0 (r ) as simply the spatial distribution in the rest of this paper. The spatial distribution is linearly related to the distribution of the EM absorption coefcient (r ) p0 (r ) = (r)(r ) (2)
2

(t |rs r | /c) dr . 4 |r s r |

(3)

where V dr represents integration over a volume of space in R3 . The time domain solution is used in several reconstruction algorithms [4], [5]. Moreover, the Fourier transform of the measured acoustic wave over time t can be written as p(r s , k ) = ikc p0 (r )G(k ; r s , r) dr.
V

(4)

with i 1, the wavenumber k = 2f /c and f denotes the frequency. This Fourier transform was used in photoacoustic image reconstruction methods described in [6], [7]. A. Waveeld Decomposition In the previous section the Greens function was mentioned and this function in R3 using the standard spherical coordinate system [3] is dened as G(k ; rs , r) eik|r s r| 4 |r s r | (5)

with sensor position r s (rs , s ), source position r (r, ), s (s , s ) and (, ). The Greens function (5) gives the pressure at rs due to monochromatic spherical, acoustic wave emanating from a point source at r . Next, we introduce the waveeld or modal decomposition of the Green,s function as a sum of orthogonal basis functions, G(k ; r s , r ) =
n,m m m Rn (k ; rs , r)Yn (s )Yn ()

(6)

where (r ) is the optical uence distribution and is the dimensionless Gr uneisen coefcient which gives a measure of the fraction of EM energy converted to stress. By solving for p0 (r ), differences in the EM absorption property of the target volume can be observed. This can identify different types of tissues or cancerous regions which have different EM absorption coefcients.
S. M. A. Salehin and T. D. Abhayapala are with the Applied Signal Processing Group, College of Engineering and Computer Science, Australian National University, Canberra, Australia. (email: asalehin@rsise.anu.edu.au; thushara.abhayapala@anu.edu.au) S. M. A. Salehin is also associated with National ICT Australia, Canberra, Australia. National ICT Australia is funded through the Australian Governments Backing Australias Ability initiative and in part through the Australian Research Council.

where Rn (k ; rs , r) = ikh(1) n (krs )jn (kr)

n

(7)

with
n,m

n=0 m=n

, n as the order, m as the mode, hn ()

and jn () are the spherical Hankel function of the rst kind and the spherical Bessel function, respectively, and denotes the complex conjugate operator. The spherical harmonics are m denoted by Yn () and are dened as
m Yn ()

(2n + 1) (n m)! m P (cos)eim 4 (n + m)! n

(8)

m where the polynomial function Pn () is the associated Legendre function.

Fig. 1. The problem involves estimating the spatial distribution enclosed in a spherical region of radius r0 from measurements on a continuous, spherical aperture, at a radius rs , completely enclosing the bounding region.

B. Fourier Transform on the 2-Sphere The spherical Harmonic function form a complete orthonormal basis function on the 2-sphere S2 with respect to the natural inner product [14]. Provided that the aperture response is square integrable on the sphere i.e. p(, k ) L2 (S2 ), the spherical harmonics decomposition [15], or the spherical Fourier transform [16] analysis and synthesis equations of the aperture response function at wavenumber k is, respectively pn,m (k ) =
s S2 m p(s , k )Yn (s ) ds

This decomposition introduced here will be used for deriving our proposed algorithm and was used previously in ultrasound reectivity imaging [8], further the algorithm presented in this work [8] was modied for photoacoustic imaging [5], [6]. This modal decomposition has the advantage of separating components dependent on the source position form those dependent on the sensor position and so found application in direction of arrival (DOA) estimation [9] and in biomedical acoustic source localization within circular sensor arrays [10], [11]. III. F REQUENCY-R ADIAL D UALITY BASED I MAGE R ECONSTRUCTION This section introduces our novel algorithm for certain ideal conditions. Extension for practical application is considered in the next section. The next subsection elaborates the problem statement, before the novel spherical Fourier-Bessel expansion of the spatial distribution is mentioned. The next two subsections highlight the two components of our proposed method for photoacoustic tomography. The last section will describe previous approaches mentioned in literature pertaining to photoacoustic tomography. Finally, the method proposed in this paper will be compared with these prior methods. A. Problem Statement We develop our algorithm by considering a hypothetical continuous aperture on the 2-sphere S2 at a radius of rs . The function p(s ) represents the sound pressure angular position measured by the ultrasound transducers. Exact relationships between the spatial distribution function and the aperture response can be preserved by using the continuous aperture assumption. Further, spatial aliasing can be completely ignored. In practical situations, there is noise present in the sensor recordings. To simplify notation and clarify our approach, we will assume ideal measurements with zero noise power in this section. The concept of using a continuous aperture to develop algorithms before an approximation is derived for discrete apertures is not new. Ward et. al. introduced the concept of the continuous aperture in [12]. Furthermore, the continuous aperture concept was applied in broadband beamforming methods [9], [13]. The problem of photoacoustic image reconstruction is an inverse problem requiring the estimation of the spatial distribution p0 (r, ) (specifying po (r ) in expression (4) in a standard spherical co-ordinate system) from measurements p(s , k ) (specifying the measured pressure due to acoustic waves p(r s , k ) on a spherical manifold at rs ). This paper considers the case where the sensor aperture is specied by samples on the sphere S2 completely enclosing the spatial distribution, and Fig. 1. illustrates the geometry of the problem formulation with the spatial distribution being zero at all spatial points with radii larger then r0 .

(9)

and p(s , k ) =
n,m 2 m pn,m (k )Yn (s )

(10)

where s S2 ds sin s ds ds and represents 0 0 integration over the sphere. The spherical Fourier coefcients pn,m are complex terms and equality in expression (10) means convergence in the mean, however if the aperture function is sufciently smooth, point-wise convergence can be assumed. The spherical Fourier transform introduced in this section will be applied in later parts of this paper. C. 3D Spherical Fourier-Bessel Expansion of Spatial Distribution D. Modal-Order Filtering of the Spatial Distribution E. Frequency-Radial Duality F. Algorithm Optimization IV. E XTENSION V. C OMPARISON
TO

D ISCRETE A PERTURES

A. Summary of the Proposed Algorithm

WITH

P REVIOUS R ECONSTRUCTION M ETHODS

A. Frequency Domain Algorithm B. Complexity Analysis VI. N UMERICAL S IMULATIONS VII. C ONCLUSION R EFERENCES
[1] G. J. Diebold, T. Sun, and M. I. Khan, Photoacoustic monopole radiation in one, two, and three dimensions, Phys. Rev. Lett., vol. 67, no. 24, pp. 33843387, Dec 1991. [2] A. C. Tam, Applications of photoacoustic sensing techniques, Rev. Mod. Phys., vol. 58, no. 2, pp. 381431, Apr 1986. [3] G. B. Arfken and H. J. Weber, Mathematical Methods for Physicists. San Diego, CA: Academic, 1995. [4] P. Liu, The p-transform and photoacoustic image reconstruction, Phys. Med. Biol., vol. 43, no. 667, 1998. [5] M. Xu, Y. Xu, and L. Wang, Time-domain reconstruction algorithms and numerical simulations for thermoacoustic tomography in various geometries, Biomedical Engineering, IEEE Transactions on, vol. 50, no. 9, pp. 1086 1099, sept. 2003. [6] Y. Xu, D. Feng, and L. Wang, Exact frequency-domain reconstruction for thermoacoustic tomography. i. planar geometry, Medical Imaging, IEEE Transactions on, vol. 21, no. 7, pp. 823 828, july 2002. [7] K. P. K ostli and P. C. Beard, Two-dimensional photoacoustic imaging by use of fourier-transform image reconstruction and a detector with an anisotropic response, Appl. Opt., vol. 42, no. 10, pp. 18991908, 2003. [Online]. Available: http://ao.osa.org/abstract.cfm?URI=ao-42-10-1899 [8] S. J. Norton and M. Linzer, Ultrasonic reectivity imaging in three dimensions: Exact inverse scattering solution for plane, cylindrical and spherical apertures, Biomedical Engineering, IEEE Transactions on, vol. BME-28, no. 2, pp. 202220, 1981.

[9] H. Teutsch and W. Kellermann, Acoustic source detection and localization based on waveeld decomposition using circular microphone arrays, The Journal of the Acoustical Society of America, vol. 120, no. 5, pp. 27242736, 2006. [Online]. Available: http://link.aip.org/link/?JAS/120/2724/1 [10] S. M. A. Salehin and T. D. Abhayapala, Localizing lung sounds: Eigen basis decomposition for localizing sources within a circular array of sensors, Journal of Signal Processing Systems, pp. 117, 2009, 10.1007/s11265-009-0435-3. [Online]. Available: http://dx.doi.org/10.1007/s11265-009-0435-3 [11] , Lung sound localization using array of acoustic sensors, in Signal Processing and Communication Systems, 2008. ICSPCS 2008. 2nd International Conference on, 2008, pp. 1 5. [12] D. B. Ward, R. A. Kennedy, and R. C. Williamson, Theory and design of broadband sensor arrays with frequency invariant far-eld beam patterns, The Journal of the Acoustical Society of America, vol. 97, no. 2, pp. 10231034, 1995. [Online]. Available: http://link.aip.org/link/?JAS/97/1023/1 [13] T. D. Abhayapala, R. A. Kennedy, and R. C. Williamson, Neareld broadband array design using a radially invariant modal expansion, The Journal of the Acoustical Society of America, vol. 107, no. 1, pp. 392 403, 2000. [Online]. Available: http://link.aip.org/link/?JAS/107/392/1 [14] D. Colton and R. Kress, Inverse Acoustic and Electromagnetic Scattering Theory, 2nd ed. Berlin, Germany: Springer-Verlag, 1998. [15] E. G. Williams, Fourier Acoustics: Sound Radiation and Neareld Acoustical Holography. New York: Academic, 1999. [16] J. R. Driscoll and D. M. Healy, Jr., Computing fourier transforms and convolutions on the 2-sphere, Adv. Appl. Math., vol. 15, no. 2, pp. 202250, 1994.