Feature Extraction and Classification of Blood Cells Using Artificial Neural Network

Transcription

1 Aerican Journal of Applied Sciences 9 (5): 65-69, 0 ISSN Science Publications Feature Extraction and Classification of Blood Cells Using Artificial Neural Network Magudeeswaran Veluchay, Karthikeyan Perual and Thiruurugan Ponuchay Departent of Electronics and Counication Enginering, PSNA College of Engineering and Technology, Dindigul, Tail nadu, India Abstract: Proble stateent: One ethod of evaluating the clinical status is counting of cell types based on features that it contains. There is a need for a rapid, reproducible ethod, superior to huan inspection and for the classification of cells. For solving these probles, quantitative digital-iage analysis is applied and a novel ethod for classifications of affected blood cells fro noral in an iage of a icroscopic section is presented. These blood cell iages are acquired fro different patient with sickle cell aneia, sickle cell disease and noral volunteers. Approach: The segentation of blood cells is ade by orphological operations such as thresholding, erosion and dilation to preserve shape and size characteristics. These features are extracted fro segented blood cells by estiating first, second order gray level statistics and algebraic oent invariants. In addition geoetrical paraeters are also coputed. The analysis of extracted features is ade to quantify their potential discriination capability of blood cells as noral and abnoral. The results obtained prove that these features are highly significant and can be used for classification. In addition, we use back propagation neural network to classify the blood cells ore efficiently. Results: For testing purposes, different sizes and various types of icroscopic blood cell iages were used and the classification efficiency is 80% and 66.6% for noral and abnoral respectively. Conclusion: The proposed syste has a good experiental result and can be applied to build an aiding syste for pathologist. Key words: Iage segentation, orphological operations, feature extraction, iage classification, artificial neural network INTRODUCTION One ethod of evaluating clinical status is the counting of cell types based on statistical and oent invariant analysis. Cells are classified as noral and abnoral (Wheeless et al., 994). Quantitative iage analysis appears to offer sensitive and reproducible classification of blood cells. Quantitative digital iage analysis has been previously applied to the study and classification of red blood cells (Bacus and Weens, 977; Bacus, 984; Gonzalez and Woods, 009; Hasani et al., 006). In a series of publications, Bacus and colleagues docuented the utility of the technique in classifying erythrocytes fro patients with various heatologic disorders. Westeran and Bacus classified cells in blood fro sickle patients into 4 different classes based on six features: size (area), heoglobin content (integrated optical density), central pallor, circularity, elongation and spicularity. This study reports the statistical and oent invariant features in the classification of cells fro blood as noral and abnoral. This approach can be divided into several Fig. : General structure of ethod After iage collection, the iage is first segented in order to isolate the interesting parts and reove noise and undesired coponents. Next, the feature extraction process is applied, to extract the well-defined stages, presented in Fig.. useful inforation fro the segented blood cells and Corresponding Author: Magudeeswaran, V., Departent of Electronics and Counication, PSNA College of Engg and Tech, Dindigul 65

2 A. J. Applied Sci., 9 (5): 65-69, 0 finally the classification can be operated according to the feature extracted by the previous stage. In addition, we use back propagation neural network to classify ore efficiently. In this work MBPN is used to identify the given class of input iage as noral and abnoral. A nuber of hidden units, learning rates, oentu and Mean Square Error (MSE) are investigated to achieve the best results. MATERIALS AND METHODS Iage Data Collection: The blood speciens were obtained fro different patients with sickle cell aneia, sickle cell disease and noral volunteers. Each blood cell iage contains nuber of noral and abnoral cells. Blood Cell Segentation: Iage segentation is used to detect the entire blood cells (Dougherty, 994; Wroblewska et al., 003). In a segented iage, the picture eleents are no longer the pixels, but connected set of pixels, all belonging to the sae region. An object can be easily detected in an iage if the object has sufficient contrast fro the background. We use edge detection and basic orphology tools to detect a cell. The individual cells are close to each other and the borders aong the are not well defined. The orphological operations ai at extracting relevant structures of the iage by probing the iage with another set of a known shape called structuring eleent, chosen as the result of prior knowledge concerning the geoetry of the relevant and irrelevant iage structures. The ost known orphological operations include erosion, dilation, opening and closing. The orphological approach to iage segentation cobines regions growing and edge detection techniques (Serra, 984; Ponsen et al., 009). The applied procedure of the iage segentation and cell separation consists of the following stages: Transforation of the original iage into gray scale Detect the entire cell using edge detection technique Application of dilation and erosion operations to sooth the object and to eliinate the distortions Feature Extraction: Twenty seven features were extracted fro each cell iage (Table ). This included 4 geoetrical features, 6 statistical features and 7 oent invariant features (Osowski et al., 004; Santinelli et al., 00). Geoetrical Features Description: We use the following geoetrical features to study characteristics of the cells: Area A-the nuber of pixels on the interior of the cell Perieter P-the total distance between consecutive points of the border Copactness C-given by the forula: perieter /area For factor F-4*3.4*Area/Perieter Table : Quantitative validation of extracted features fro blood cell iages Noral cell Abnoral cell Feature Mean SD Mean SD T-Test (P X) Area Perieter E-050 Circularity For factor E-050 K (No of pixels) E-060 Mean Dispersion Variance Average Energy E-050 Skew ness Kurtosis Median Mode Energy Inertia Entropy Hoogeneity Max probability Inverse Correlation Φ E-080 Φ E-080 Φ Φ E-070 Φ E-070 Φ E-060 Φ E

3 A. J. Applied Sci., 9 (5): 65-69, 0 First order statistical features of the iage: The next set of features has been generated on the basis of the intensity distribution of the iage. If u (, n) is the discrete iage and is the total nuber of pixels in cell region. The first order gray level statistical features ean (M), dispersion (M), variance (M3), average energy (M4), Skew ness (M5), kurtosis (M6), edian (M7) and ode (M8) are estiated. The first order gray level statistical feature forulas are given by Eq. -6: M(k,l) = [u( k, n l)] () n R M(k,l) = u( k, n l) M(k,l) () n R M3(k, l) = [u( k, n l) M(k, l)] (3) n R M4(k,l) = [u( k, n l)] (4) n R M5(k,l) = [u( k, n l) M(k,l)] (5) n R 3 E P (i, j) = kl N (9) H = N P kl(i, j)log[p kl(i, j)] (0) C = (i j)p (i, j) µ µ N kl xx y () σxσy () I R = (i j) P kl (i, j) N L = P (i, j) N () kl (3) + Algebraic Moent Invariants: The classic technique for generating Invariants in ters of algebraic oents. If u (, n) is discrete input iage, the algebraic oents is given by Eq. 4: p q = n u(,n) (4) n M6(k,l) = [u( k, n l) M(k,l)] (6) n R 4 If the intensity values I (k, l) of u (, n) are arranged in ascending order, then Eq. 7: N + M7 k,l I k, l of pixel th ( ) = ( ) (7) In case of even N, the edian value is estiated by finding the average of two iddle values. If is the histogra of the iage that gives the nuber of pixels with gray levels, then Eq. 8: M8(k, l) =I(k, l)of ax{h(x i )} (8) Second Order Statistical Features of the Iage: These statistical features of second order are coputed in a two step process. The first step delivers the Cooccurrence atrices containing the eleent. Each (i,j)th entry of the atrices represents the probability of going fro pixel with gray level (i) to another with a gray level (j) under a predefined angles, 0, 30, 60, 90, 0 and 50. Based on the co-occurrence atrices texture features are estiated. Five texture features energy (E), entropy (H), correlation (C), inertia (In) and hoogeneity (L) are calculated as given by Eq. 9-3: Here a fixed nuber of lower order oents for each pixel in the iage are coputed. Usually oent invariants are specified in ters of noralized central oents, which is given by Eq. 5: µ η = (5) µ Where: γ oo p q ( ) (n n) u(,n) n µ = = n = p + q γ = + forp + q =,3,... Using above Equation for noralized central oents, a set of seven oent invariant features φ = [φ, φ,. φ 7 ] is calculated for the blood cell iages. Back Propagation Neural Network (BPNN): A back propagation neural network was created to classify the iages (Markiewiczl and Osowski, 006). 67

4 A. J. Applied Sci., 9 (5): 65-69, 0 Fig. 3: Blood cell iage Fig. : Architecture of a neural network Table : MBPN learning paraeters and Classification rates for two classes of blood cell iages Iage No of input No of hidden No of output Classification category neurons neurons neurons efficiency (%) Noral Abnoral Fig. shows the architecture of a neural network that contains nuber of input, output and hidden layers. The nuber of neurons in the input layer was fixed as 7, nuber of neurons in the hidden layer as and the nuber of neurons in the output layer as. The paraeters oentu and learning rate was set as 0.00 and 0.3 respectively, for training the network. The training function used was traingda. The axiu nuber of epochs was set as The activation function was initially set as tansig and in the hidden layer and purelin at the output layer. All these networklearning paraeters are suitably optiized to achieve better classification efficiency. Fig. 4: Segentation cell iage (a) (b) RESULTS Fig. 5: Blood cells of noral The quantitative easures of the derived 7 features are enuerated in Table. The standard deviation and ean value of each feature is deterined and it has been given for all blood cell classes. The recognition and classification of cells have been perfored for different blood cell classes. They include noral, abnoral cells. Table presents MBPN learning paraeters and Classification rates for two classes of blood cell iages. The Fig. 3 shows blood cell iage. Fig. 4 shows the pre-processed blood cell iage that has well defined shape and structure. (a) Fig. 5 presents noral blood cells. Fig. 6 presents abnoral blood cells. Fig. 6: Blood cells of abnoral 68 (b)

5 DISCUSSION Our results indicate that iage analysis ay be used for the autoatic classification of red blood cells fro patients with sickle cell aneia (SS) and SC disease into the pattern classes of noral and other abnoral. The potential of feature in identifying the category is verified statistically by student t-test to evaluate p value. The result attained for first order gray level statistical features shows that the features K (No of Pixels), ean, edian, average energy and ode are very highly significant (p E-06 for K; p for Mean; p 9.53E-05 for average energy p for Median; p for Mode) in discriinating the cell category which shows 95% significant difference in ean values. This followed by skew-ness is highly significant (which shows 90% significant difference) and kurtosis, dispersion and variance are less significant features (85%). The study on features obtained by second order gray level statistical features reveals that the features energy and inverse are highly significant (p , ). The reaining features of this ethod are less significant. In case of algebraic oent invariant features ф, ф 3, ф 4, ф 5, ф 6, ф 7 perfor well with a significant level of very high and reaining features of this ethod are less significant. In addition geoetrical features perieter, for factor are very high significant which shows 95% significant difference and reaining are highly significant. The MPBN is trained with derived paraeter irrespective of their significant level and association within the category. The network paraeters, selection of nuber of hidden layer and hidden layer neurons are optiized to yield better classification efficiency. This provides exact classification efficiency of 80% and 66.6% for noral and abnoral respectively. CONCLUSION The study presents an approach for classification of blood cells as noral and abnoral in icroscopic iages using statistical and oent invariant analysis. Obtained results indicate that soe specific features are highly significant in classification and back propagation neural network is also trained to enhance the process of classification. The network is test unknown blood cell iages. The classification provided by network is appreciable. This also helps the edical technician to objectively decide on the nature of blood. ACKNOWLEDGMENT A. J. Applied Sci., 9 (5): 65-69, 0 The researchers would like to thank all the volunteers who had participated in this study and the faculties of PSNA College of Engineering and 69 Technology, Dindigul for their valuable help and support to coplete this study. REFERENCES Bacus, J.W. and J.H. Weens, 977. An autoated ethod of differential red blood cell classification with application to the diagnosis of aneia. J. Histoche. Cytoche., 5: DOI: 0.77/ Bacus, J.W., 984. Quantitative red cell orphology. Monogr. Clin. Cytol., 9: l-7. PMID: Dougherty, E.R., 994. Digital Iage Processing Methods. st Edn., Dekker, New York, ISBN-0: X pp: 47. Gonzalez, R.C., R.E. Woods, 009. Digital Iage Processing. 3rd Edn., Pearson Prentice Hall, India, ISBN-0: pp: 954. Hasani, J.S.A., I.M.M.E. Eary, H.N. Heyasat and A.A. Al-zu'bi, 006. On the application of artificial neural networks in analyzing and classifying the huan chroosoes. Al- Balqa' Applied University. Markiewiczl, T. and S. Osowski, 006. Data ining techniques for feature selection in blood cell recognition. Proceedings of the European Syposiu on Artificial Neural Networks, Apr. 6-8, Bruges, Belgiu, pp: Osowski, S., T. Markiewicz, B. Marianska and L. Moszczynski, 004. Feature generation for the cell iage recognition of yelogenous leukeia. Warsaw University Technology. Ponsen, S., N.A. Narkkong, S. Paok and W. Aengwanich, 009. Coparative heatological values, orphoetric and orphological observation of the blood cell in capture and culture asian eel, onopterus albus (Zuiew). A. J. Ani. Vet. Sci., 4: DOI: /ajavsp Santinelli, A., R. Mazzucchelli, P. Colanzi, A. Tinca and R. Montironi, 00. Iage processing, diagnostic inforation extraction and quantitative assessent in pathology. J. Cell Mol. Med., 6: DOI: 0./j tb0034.x Serra, J., 984. Iage Analysis and Matheatical Morphology. st Edn., Acadeic, London, ISBN- 0: 06374X, pp: 60. Wheeless, L.L., R.D. Robinson, O.P. Lapets, C. Cox and A. Rubio et al., 994. Classification of red blood cells as noral, sickle, or other abnoral, using a single iage analysis feature. Cytoetry, 7: DOI: 0.00/cyto Wroblewska, A., P. Boninski, A. Przelaskowski and M. Kazubek, 003. Segentation and feature extraction for reliable classification of icrocalcifications in digital aogras. Instit. Radio Elect., : 7-35.