Ultrasound Kidney Image Analysis

Determination of Kidney Area Independent Unconstrained Features for Automated Diagnosis and Classification : 

Determination of Kidney Area Independent Unconstrained Features for Automated Diagnosis and Classification K. Bommanna Raja and M. Madheswaran International Conference on Intelligence and Advance Systems-2007

Objectives : 

Nov. 25 - 28 2 Objectives To evaluate the tissue characteristic of kidney for implementing unbiased diagnosis procedure and to classify important kidney orders To establish a set of unconstraint features that are independent to kidney area variations

Need… : 

Nov. 25 - 28 3 Need… Realizing CAD system exclusively for US kidney image is made practicable. With such system it is possible to (i). Establish a quantitative universal reference for the US kidney images (ii). Implement image retrieval in medical application (IRMA) system (iii). Make comparative study on images for better decision making (iv). Develop an expert system that automatically recognizes the extent of pathology or normality (v). Examine extent of healing or failure under post-therapy observation

Sample US kidney Images : 

Nov. 25 - 28 4 Sample US kidney Images

Material and Methods : 

Nov. 25 - 28 5 Material and Methods Image Data Collection Two types of scanning systems namely ATL HDI 5000 curvilinear probe with transducer frequency of 5 – 6 MHz and WiproGE LOGIC 400 curvilinear probe with transducer frequency of 3 – 5 MHz. The longitudinal cross section of the kidney is taken by fixing the transducer frequency at 4 MHz. In each class 50 images are obtained. In total 150 images are pre-processed before feature extraction. The necessary care has been taken to preserve the shape, size and gray-level distribution as it obliterates the sonographic content of information.

Material and Methods : 

Nov. 25 - 28 6 Material and Methods Image Pre-processing Segmentation by higher order spline interpolation after up-sampling of distributed coordinate Rotation to zero degree axis Retaining the pixel of interest Estimation of Content Descriptive Features Kidney Characterization

Material and Methods : 

7 Material and Methods Image Pre-processing Input US kidney image i-HSIC segmentation Image rotation to zero degree reference axis Unbounded pixel elimination Nov. 25 - 28

Material and Methods : 

Nov. 25 - 28 8 Material and Methods Feature Extraction First order gray level statistical features Second order gray level statistical features Algebraic moment invariants features Multi-scale differential features Power spectral features Dominant Gabor wavelet features

Material and Methods : 

9 Material and Methods Feature Extraction First order gray level statistical features mean (M1), dispersion (M2), variance (M3), average energy (M4), skewness (M5), kurtosis (M6), median (M7) and mode (M8) Second order gray level statistical features energy (E), entropy (H), correlation (C), inertia (In) and homogeneity (L) Algebraic moment invariants features eight RST invariant features ф1, ф2, ф3, ф4, ф5, ф6, ф7 and ф5/ф1 Nov. 25 - 28

Material and Methods : 

10 Material and Methods Feature Extraction Multi-scale differential features two principal curvature features namely isophote (N) and flowline (T) are computed. From these values of N and T, a set of MSDF’s are then determined, namely, the mean (Nmean; Tmean), maximum (Nmax; Tmax) and minimum (Nmin; Tmin) Power spectral features six power spectral features denoted by and are estimated at the specific cut-off frequencies in the spectrum and by considering global mean total power. Dominant Gabor wavelet features Out of 30 Gabor wavelets, a unique Dominant Gabor Wavelet is determined by estimating the similarity metrics between original and reconstructed Gabor image. The Gabor features ‘μmn’, ‘σmn’ and ‘AADmn’ are then evaluated using Dominant Gabor Wavelet Nov. 25 - 28

Decision Support System For Kidney Classification : 

11 Decision Support System For Kidney Classification . . . . Input feature vector Ij Fuzzification fj All 36 If X≥n/2 Yes NR No Initiate Optimized MBPN MRD CC Fuzzy rules FIS Hybrid fuzzy-neural system Nov. 25 - 28

Decision Support System For Kidney Classification : 

12 Decision Support System For Kidney Classification WHILE Input- testdata FUZZIFY the testdata INITIALIZE a variable NORM_Count to zero DESIGN inference rules FOR all inference rules, IF rule is true THEN increment NORM_Count by 1 END IF FIX x to be a number not less than half the total number of rules IF NORM_Count>=x THEN display ‘NR’ ELSE SIMULATE the trained MBPN with the test data DISPLAY ‘MRD’ or ‘CC’ based on the simulated output END IF END WHILE Nov. 25 - 28

Decision Support System For Kidney Classification : 

13 Decision Support System For Kidney Classification The imprecision of the input feature values are defined by fuzzy numbers that lies between 1 and 10 via normalization. The fuzzy sets are defined with fuzzy number expressed by linguistic variables minimum (A), medium (B) and maximum (C). The fuzzy set ‘A’ includes fuzzy numbers in the range 0 – 3.35, ‘B’ contains 3.36 – 6.65 and ‘C’ consists of 6.66 – 10. The fuzzy rules for all 36 features are linguistic and of if – then constructions that have the general form, Nov. 25 - 28

Decision Support System For Kidney Classification : 

14 Decision Support System For Kidney Classification if [(fij is A) and (fij is B)] or [(fij is B) and (fij is C)] or [(fij is A) and (fij is C)] then Cp . . . . if [Cp is X] then NR Here ‘i’ indicates rule, ‘j’ refers to feature, ‘Cp’ is the count and ‘X’ is the number of count. If Cp is reaches the value X, the fuzzy system outputs the category as NR Details of Fuzzy Rules and Count Set Nov. 25 - 28

Decision Support System For Kidney Classification : 

15 Decision Support System For Kidney Classification Detail of MBPN architecture parameters fixed for Fuzzy – neural system Nov. 25 - 28

Verification For Unconstraint Behavior : 

16 Verification For Unconstraint Behavior The classification efficiency achieved with proposed fuzzy-neural system is appreciable, But the important issue that has to be investigated is to study the dependency of features on kidney area Longitudinal cross section of kidney obtained for subjects varies due to variation in position and orientation of transducer probe placed on the body surface As the feature values depend on the gray level intensity distribution of pixels, the change in kidney area may influence the feature value and hence mislead decision making Requirement of the CAD system is, the estimated features must be independent of kidney area and unconstraint in making decision regarding the kidney category. Nov. 25 - 28

Verification For Unconstraint Behavior : 

17 Verification For Unconstraint Behavior Pre-processed US kidney images showing variation in kidney area. a. normal male subject with age 38 years b. normal male subject with age 43 years c. MRD male subject with age 54 years d. MRD female subject with age 61 years e. CC male subject with age 49 years f. CC female subject with age 57 years Nov. 25 - 28

Verification For Unconstraint Behavior : 

Nov. 25 - 28 18 Verification For Unconstraint Behavior In general the kidney area varies between : 15256 and 33246 for NR images 11434 and 21112 for MRD 16541 and 28943 for CC

Verification For Unconstraint Behavior : 

Nov. 25 - 28 19 Verification For Unconstraint Behavior Three statistical measures are used to verify the dependency F-test is performed which returns the one-tailed probability ‘p’ which notify that the variances of one data set and other are not significantly different Pearson product moment correlation coefficient ‘rP’ is estimated to understand the extent of linear relationship between kidney area and a feature Sixth order polynomial regression analysis is performed to calculate the R2 - value between two data sets

Results : 

20 Results Classification efficiency obtained with MBPN-ANN and Fuzzy-Neural system for 36 features Nov. 25 - 28

Results : 

21 Results Statistical measure for the features to verify unconstraint behavior with respect to kidney area Nov. 25 - 28

Results : 

22 Results Statistical measure for the features to verify unconstraint behavior with respect to kidney area Nov. 25 - 28

Results : 

23 Results P P Statistical measure for the features to verify unconstraint behavior with respect to kidney area Nov. 25 - 28

Conclusion : 

24 Conclusion A reliable method for diagnosing and classifying the US kidney images is developed and implemented using first order statistical and algebraic moment invariant features These features highly depict the characteristic of kidney and shows promising performance in classification of kidney images as normal, medical renal diseases and cortical cyst Most of the features are independent to kidney area variations The developed method not only helps in classification also extends its potential in realization of automated CAD system. Nov. 25 - 28

References : 

25 References [1] Hagen, S L, 4th eds. 1995. Urinary System, In: Diagnostic Ultrasonagraphy. Reading: St.Louis Mosby. [2] Anant Madabhushi.; and Dimitris N.; and Mctaxas. 2003. Combining Low-, High-Level and empirical Domain Knowledge for Automated Segmentation of Ultrasonic Breast Lesions. IEEE Trans. on Medical Imaging, Vol. 22, No.2: 155-169. [3] J.Ravell.; M.Mirmehdi.; and D.McNally. Applied Review of Ultrasound Image Feature Extraction Methods. 2002. In Proc. of 6th Medical Image Understanding and Analysis Conference, 173–176. A Houston and R.Zwiggelaar, Editors, BMVA Press. [4] Sheng-Fang Huang, Ruey-Feng Chang, Dar-Ren Chen and Woo Kyung Moon, “Characterization of Speculation on Ultrasound Lesions”, IEEE Trans. on Medical Imaging, Vol.23, No.1, pp. 111–121, 2004. [5] Bakker J, Olree M, Kaatee R, de Lange E.E, and Beck R.J.A, “Invitro Measurement of Kidney Size: Comparison of Ultrasonography and MRI”, Ultrasound in Med. Biol., Vol. 24, No.5, pp. 683– 688, 1997. [6] Matre K, Stokke E.M, Martens D and Gilija O.H., “Invitro Estimation of Kidneys Using 3-D Ultrasonography and a Position Sensor”, Eur. J. Ultrasound, Vol.10, pp. 65–73. 1999. [7] Marcos Martin-Fernandez and Carlos Alberola-Lopez, “An Approach for Contour Detection of Human Kidneys from Ultrasound Images using Markov Random Fields and Active Contours”, Med. Image Analysis, Vol.9, pp. 1–23, 2005. [8] Jun Xie, Yifeng Jiang and Hung-tat Tsui, “Segmentation of kidney from ultrasound images based on texture and shape priors”, IEEE Trans. on Med. Imaging, Vol.24, No.1, pp. 45–57, 2005. [9] Lin D.T., Lei C.C. and Hund S.W., “Computer-Aided Kidney Segmentation on Abdominal CT Images”, IEEE Trans. on Inf. Tech. in Biomed., Vol.10, No.1, pp. 59–65, 2006. [10] A.Eslami, M.Jahed and M.Naroienejad, “Fully Automated Cyst Segmentation in Ultrasound Image of Kidney”, Proc. 3rd IASTED Intl. Conf. on Biomedical Engineering, Austria, PaperID -19418, 2005. Nov. 25 - 28

References : 

26 References [11] K.Bommanna Raja, M.Madheswaran and K.Thyagarajah, “A General Segmentation Scheme for Contouring Kidney Region in Ultrasound Kidney Images using Improved Higher Order Spline Interpolation”, Intl. J. of Biomedical Sciences, Vol. 2, No.2, pp. 81–88, 2007. [12] Ahmadian A., Mostafa A., Abolhassani M.D. and Salimpour Y, “A Texture Classification Method for Diffused Liver Diseases using Gabor Wavelets”, IEEE Proc. on 27th Annual Intl. Conf. of the Engineering in Medicine and Biology Society, pp. 1567 – 1570, 2005. [13] K.Bommanna Raja, M.Madheswaran and K.Thyagarajah, “Ultrasound Kidney Image Analysis for Computerized Disorder Identification and Classification using Content Descriptive Power Spectral Features”, Journal of Medical Systems, Vol. 31, pp. 307–317, 2007. [14] K.Bommanna Raja, M.Madheswaran and K.Thyagarajah, “Evaluation of Tissue Characteristics of Kidney for Diagnosis and Classification using First Order Statistics and RTS invariants”, IEEE Proc. of Intl. Conf. on Signal Processing, Comm. and Networking, MIT, Chennai, Vol. 1, pp. 483–487, 2007. [15] K.Bommanna Raja, M.Madheswaran and K. Thyagarajah, “Analysis of Ultrasound Kidney Images using Content Descriptive Multiple Features for Disorder Identification and ANN based Classification”, IEEE Proc. of Intl. Conf. on Computing: Theory and Applications, Indian Statistical Institute, Kolkatta, Vol. 1, pp. 382–388, 2007. [16] K. Bommanna Raja, M.Madheswaran and K. Thyagarajah, “Quantitative and Qualitative Evaluation of US Kidney Images for Disorder Classification using Multi-scale Differential Features”, ICGST Intl. J. on Bio Informatics and Medical Engg., Vol. 7, No. 1, pp. 1–8, 2007. [17] K.Bommanna Raja, M.Madheswaran and K.Thyagarajah, “Analysis of Ultrasound Kidney Images for Disorder Identification and Classification using Dominant Gabor Wavelet (DoM-GW)”, Machine Vision and Applications Journal, Manuscript ID MVA-Apr-07-0060, Communicated, 2007. [18] K.Bommanna Raja, M.Madheswaran and K.Thyagarajah, “A Hybrid Fuzzy-Neural System for Computer-Aided Diagnosis of Ultrasound Kidney Images using Prominent Features”, Journal of Medical Systems, Manuscript ID JOMS 135R1, accepted, available on-line, 2007. Nov. 25 - 28

Slide 27: 

27 Thank You K.Bommanna Raja Centre for Research and Development Department of Electronics and Communication Engineering PSNA College of Engineering and Technology Dindigul – 624 622, Tamil Nadu, INDIA. Tel:91-451-2554417, E-mail:bommanna_raja@yahoo.com Tel:91-451-2554262, E-mail:contact@psnacet.org M.Madheswaran Centre for Advanced Research Department of Electronics and Communication Engineering Muthayammal College of Engineering Rasipuram – 637 408, Tamil Nadu, INDIA. Tel:91-4287-22683, E-mail: madhi_eswaran@yahoo.co.in Nov. 25 - 28