Working Group 2: Texture Formation
1. Texture Analysis of Cadaverous Kidney
Texture Parameters of kidneys from human cadavers IN First Hours after the Removal
M.Hájek , M.Babiš , V.Herynek , P.Táborský*, E.Pokorná**, Š.Vítko**, Michal Strelecki
Prague and Lodz
Texture analysis was applied for the evaluatio of T2 maps and T2w images
of kidneys from human cadavers. T2 relaxation times of 54 human
cadaverous kidneys were measured by using 16 points CGPM sequence in different
times after the removal together with T2 weighted images. Relaxation
T2 maps were evaluated as histograms and first order texture
parameters Mean, variation, skewness and kurtosis
were used for the quantitative comparison of the kidney in time. First
and second order texture analysis parameters of T2w images were used for
the description of cadaverous kidney too. The mean change of the first-order
statistical properties of T2 histogram of human kidney between
4th and 8th hour after the removal was found smaller
compared to pigs kidney model. We did not find in this study of human cadaverous
kidneys any significant changes of relaxation time T2 and texture
parameters during cold ischemia at least up to 8 hours after the removal.
2. Texture Analysis of Human Liver
Daniel Jirák, Monika Dezortová, Pavel Taimr*, Milan Hájek
An attempt is made to try automatic classification of MR images of human cirrhotic liver by using Texture Analysis (TA). For any image texture, a large number of its statistical features can be computed. The aim of the study was: 1) eliminate mistakes of classifying of healthy and disordered liver by TA and help determine clinical stage in unclear cases; 2) to find features which describe the texture of MR images of the human liver by the most discriminative power.
We measured patients (39) and controls (10) at 1.5 T imager (Siemens Vision). We used these MRI sequences: T2w BH (Breath Hold) transversal slices (8 mm, TR/TE=4200/138 ms, 300 FOV, 350 FOV), 1 ACQ (BH), 10 ACQ. Statistical features were computed by program Mazda (1) and by using programs Convert and B11 were made the classification. For our study we used 137 images. We divided patients to the three groups (A,B,C) according to their clinical stage described by Child Pugh Score (CPS ascites, protrombin time, bilirubin level, albumin level (in the blood), degree of hepatic encephalopathy). Into each group we added data of controls. All data were tested by various methods (PCA-Principle Component Analysis, k-NN – Nearest Neighbor Classifier, etc.) For classification of all groups we used five different sets of TA parameters, each set of parameters contained minimum 3 and maximum 10 parameters.
All TA methods based on full set of images (Fig.1) proved to differentiate in each group A,B,C the data of controls and patients. The error of classification (i.g. wrong classification) depends on the set of choosing parameters (Table 1), it moved around the 8%, better results were for groups, which contained bigger number of patients. The best set of parameters includes 10 parameters of the histogram of the second order Angular Second Moment, Contrast, Correlation, Entropy, Inverse Difference Moment, Sum Average, Sum Variance, Sum Entropy, Difference Variance, Difference Entropy). Very good result brought the using the sets with the parameters of the histogram of the first order. TA based on means of TA parameters of each patient and control wasn’t accurate, the classification errors were around 45%. The classification of randomly chosen patient was always found positive (we repeated it 10 times).
Discussion: We proved that the texture analysis can be successfully used for the separation of cirrhotic patient and healthy volunteers. Different sets of TA parameters can be used for similar classification of patient. From our tested parameters the most significant power have kurtosis and difference entropy. We suppose that increasing of training set of patient and controls will improve classification power of the TA.
 http://www.eletel.p.lodz.pl/cost/download.html Andrzej Materka – Mazda User‘s Manual, 1999
Figure 1. Example of separation of the controls (marked
Numb.1) and patients (Numb.2) by using methods PCA
|Set of TA parameters||Without std.||With std.|
Table 1. Example of Classification by using different
sets of parameters and the method Nearest Neighbor. Number of misclassification
with and without feature standartization is described.
Multiple Sclerosis and MS analysis - dataset from Basel G. Szekely (CH)
MS Strasbourg Chambron(F)
Muscle of the knee texture analysis Herlidou (F)
Muscle of the face texture analysis Mahmoud (F)
The combination of data analysis of TA and 31P MRS will be performed (Prague)
TA will be used for the description of irradiation changes in rats brain (Prague)
Abdominal Fat and Muscles