Cardio -Vascular Image Contrast Improvement via Hardware Design

Transcription

1 Cardio -Vascular Image Contrast Improvement via Hardware Design Aimé Lay-Ekuakille 1, Giuseppe Vendramin 1, Ivonne Sgura, Amerigo Trotta 3 1 Dipartimento d Ingegneria dell Innovazione, University o Salento, Via Monteroni, Lecce, 73100, Italy, aime.lay.ekuakille@unile.it, giuseppe.vendramin@unile.it, Ph Dipartimento di matematica Ennio De Giorgi,University o Salento,Via Arnesano, Lecce, 73100, Italy, ivonne.sgura@unile.it 3 Dipartimento di Elettrotecnica ed Elettronica, Polytechnic Faculty o Bari, Via Orabona 4, Bari, 70100, Italy, trotta@misure.poliba.it Abstract - In many circumstances sotware processing is the best and suitable way o improving image contrast, hence image quality. Dierent biomedical instrumentation and apparatuses encompass post processing units capable o optimal outcomes. However, in particular cases, especially or contour analysis, the above outcomes are not suicient. Thereore, an alternative can be envisaged in designing image enhancer and intensiier to compensate image processing limitations. This aspect is very important regardless o computational costs o this alternative. This paper proposes some criteria o design image enhancer and intensiier or biomedical applications. Moreover, the hardware has been used or improving the quality o cardio-vascular image constrast with acceptable outcomes. The images come rom dierent instrumentations, namely, diagnostic ultrasound and echocardiograph. In addition, a speciic algorithm has been implemented by means o a variational approach based on Mumord & Shah unctional to solve the magniication or zooming problem or a given digital image. The numerical solution o the system o elliptic PDEs associated with the MS unctional can provide simultaneously noise suppression, extraction o shape and magniication o the image. I. Introduction Practical noise reduction methods oten involve mutiple-sample averaging (block averaging) o a sequence o measured values, x(n), to compute a sequence o N-sample arithmetic means, M(q). As such, the block averaged sequence M(q) is deined by: ( q 1) N 1 x k = qn M ( q) = ( n), (1) where the time index o the averaging process is q=0,1,,3, etc. When N=10 or example, or the irst block o data (q=0), time samples x(0) through x(9) are averaged to compute M(0). For the second block o data (q=1), time samples x(10) through x(19) are averaged to compute M(1), and so on [1]. The ollowing impulsive-noise smoothing algorithm processes a block o time-domain samples, obtained through periodic sampling, and the number o samples, N, may be varied according to individual needs and processing resources. The processing o a single block o N time samples proceeds as ollows: collect N samples o x(n), discard the maximum (most positive) and minimum (most negative) samples to obtain an N-sample block o data, and compute the arithmetic mean, M(q), o the N samples. Each sample in the block is then compared to the mean. The direction o each sample relative to the mean (greater than, or less than) is accumulated, as well as the cumulative magnitude o the deviation o the samples in one direction (which, by deinition o the mean, equals that o the other direction). This data is used to compute a correction term that is added to the mean according to the ollowing ormula: A( q) ( P M ( q) os N N eg ) D total = () where A(q) is the corrected mean, M(q) is the arithmetic mean (average) rom Eq.(1), P os is the number o samples greater than M(q), and N eg is the number o samples less than M(q), and D total is the sum o deviations rom the mean (absolute values and one direction onl. D total, then, is the sum o the dierences between the P os samples and M(q).

2 The aorementioned algorithm can be combined with a morphological iltering process. The morphological ilters [] are the particular non linear ilters based on the morphological maths. In this circles every signal is the morphological convolution o itsel and a structural element. The main tools o morphological image processing are a broad class o nonlinear image operators, o which the two most undamental are dilation [3] and erosion [4]. The space domain in which images are deined can be either continuous, E = R, or discrete, E = Z. For a binary image represented by a set S E, its morphological dilation by another planar set B, denoted by S B, and its erosion, denoted by S B, are the Minkowski set operations S B = { x b : x S, b B } (1) { x B x S } where x = ( denotes points on the plane and = { x b b B} S B = : () : denotes set translation by a vector. The corresponding signal operations are the Minkowski dilation and erosion or an image unction (ig.1) B x Fig. 1 Morphological image processing main tools : E R by another (structuring) unction g : g( x) = ( g( x y E (3) g( x) = ( y x) y E (4) where and denote supremum and inimum. There is, however, a variety o nonlinear smoothing ilters that can smooth while preserving important image eatures and can provide a multi scale image ensemble, that is, a nonlinear scale-space. A large class o such ilters consists o the standard morphological openings and closings (which are serial compositions o erosions and dilations) in a multi scale ormulation [5] [6] [7] and their lattice extensions, o which the most notable are the reconstruction openings and closings [8] [9]. II. Design The preliminary employment o an ampliication section has been necessary just because o the aorementioned characteristics. Thanks to the use o the ampliier a little improvement o the echographic image (see ig. and ig.3 or acilit has been obtained. In order to get a urther important improvement, it was decided to update the processing card by designing and assembling a video ampliier-enhancer. A video enhancer, also called crispener, is a device which enhances the high

3 requencies o video signal. It does not bring to an actual enlargement o signal band, but the enhancement o the higher requencies produces a better contrast in details. The used circuit layout is shown in ig. 4. Ater assembling and testing the enhancer, it has been connected to the ampliier through the connection shown in ig. 5. Fig. Instrumentation keyboard Fig.3 Used linear probe Fig.4 Video enhancer circuit layout Fig.5 Video ampliier-enhancer circuit layout Fig. 6 Step response Fig.7 HPE4411B spectrum analyzer Fig,6 and ig.7 depict the step response o design video ampliier-enhancer circuit and the spectrum respectively. III. Results and conclusion The quality and improvement o the obtained images are very evident: in the starting image (ig.8) no object can be distinguished, whereas in ig. 9 a gland is seen. The research has traced out a way o designing an image processor implementing wavelet and nonlinear iltering as morphological one. It establishes some interesting and speciic criteria or a correct compatibility between image processing and hardware treatment. Certainly, in presence o black and white image (or gland) o echographic scanner, the best results are obtained using a morphological iltering.

4 Fig. 8 Original image rom the Combison 310 Fig. 9 Processed image via ampliier-enhancer Our idea is to use or magniication o echographic images the variational approach based on the above Mumord-Shah model which provides simultaneously and in a single ramework the image segmentation, denoising, and magniication. The numerical treatment o the above Mumord & Shah unctional is not easy or the presence o the unknown set K. By the way, the unctional MS(K; u) can be approximated in the sense o the Γ-convergence [10], Ambrosio & Tortorelli [11] by means o a sequence o elliptic unctionals Fk. These elliptic unctionals are more convenient because the associated system o Euler-Lagrange equations are more tractable rom the numerical point o view. In order to apply the MS approach both or denoising and edge detection in a given image g, the authors o [1] have introduced a inite dierence approximation o the elliptic PDE system together with an algorithm which varies the parameter k to perorm a simulation o the Γ-convergence in the discrete ramework. In this paper we aim to apply the same approach and the same numerical methods as in [1] or the magniication problem o a given noised image. Towards this goal, the MS approach can be easily applied by solving the above PDE system by simply denoising in a dierent way the so called idelity term present in the unctional, that is µ u( dxdy = µ u( dxdy (5) Ω\ D D where D is the region to be painted. In other words, this can be obtained also denoising the unction µd(x; =µ i (x; Є Ω\D and and µd(x; = 0 otherwise. The outcomes o the application o Mumord-Shah model, starting rom the original image (ig.10), are shown in ig. 11 (or one iteration) and ig.1 (ater two iterations). The MS approach can be replace the other approaches. Ω Fig. 10 Original image Fig. 11 MS - iteration 1 Fig. 1 MS - iteration

5 Reerences [1] P. Lynn, and W. Fuerst, Introductory Digital Signal Processing, with Computer Applications, John Wiley & Sons, New York, [] R.T. Rockaellar, Convex analysis,princeton University Press, [3] J. Serra, Image analysis and Mathematical Morphology, Academic Press, New York, 198. [4] J. Serra (ed.), Image analysis and Mathematical Morphology Vol., Academic Press, New York, [5] M.Chen, P.Yan, A Multiscaling Approch Based on Morphological Filtering, IEEE Trans. Pattern Anal. Machine Intell. 11, , July, [6] P.Maragos, Pattern Spectrum and Multiscale Shape Representation, IEEE Trans. Pattern Anal. Machine Intell. 11, , July, [7] G.Matheron, Random Sets and Integral Geometry, Wiley, New York, [8]P.Salembier, J.Serra, Flat Zones Filtering, Conencted Operators, and Filters by Reconstruction, IEEE Trans. Image Process. 4, , August, [9] L.Vincent, Morphological Gray Scale Reconstruction in Image Analysis: Applications and Eicient Algorithms, IEEE Trans. Image Process., , April, [10] L. Ambrosio, L. Faina & R.March: Variational approximation o a second order ree discontinuity problem in Computer Vision, SIAM J. Math. Anal. 3 (001), [11] L. Ambrosio & V.M. Tortorelli: On the approximation o ree discontinuity problems, Boll. U.M.I. 6-B (199), [1] M. Carriero, A Farina & I. Sgura: Image Segmentation in the Framework o Free Discontinuity Problems - in D. Pallara (Ed.): Calculus o Variations: Topics rom the Mathematical Heritage o Ennio DeGiorgi, Quaderni di Matematica, Vol. 14, Dipartimento di Matematica Seconda Universita' di Napoli - 004, pp