MAZDA A SOFTWARE FOR TEXTURE ANALYSIS Piotr M

MAZDA – A SOFTWARE FOR TEXTURE ANALYSIS Piotr M. Szczypinski, Michal. Strzelecki, Andrzej Materka Institute of Electronics, Technical University of Lodz, Wolczanska 211/215, 90 -924 Lodz, Poland International Symposium on Information Technology Convergence, Jeonju, Korea, 2007

Functionality A Ma. Zda: – is a software package for 2 D and 3 D image texture analysis – computes a variety of textural features within arbitrarily shaped regions of interest – computes feature maps of images – performs statistical analysis of computed feature sets – aids in image texture classification ISITC 2007, Ma. Zda - a software. . . 2

Functionality Ma. Zda has been under development since 1998, to satisfy the needs of the participants of COST B 11 European project "Quantitative Analysis of Magnetic Resonance Image Texture" and the subsequent COST B 21 "Physiological modelling of magnetic resonance image formation". The program code has been written in C++/Delphi and compiled for PC computers that use MS Windows® XP operating system. [Ma. Zda Manual] ISITC 2007, Ma. Zda - a software. . . 3

The long story short • • • development started in 1996 with the Mammo program (Łódź-Warszawa) combining procedures from NMRWin (DKFZ-Heidelberg) in 1999 b 11, a computer program for data analysis, classification, and vizualisation was developed as a complementary software to Ma. Zda 1998 -2002 development within COST B 11, 2004 -2007 COST B 21 project aimed at analysis of magnetic resonance images texture The name of the program is an acronym derived from. Macierz ‘ Zdarzen’ that is Polish counterpart of the English term ‘co-occurrence matrix’. ISITC 2007, Ma. Zda - a software. . . 4

Texture – – perceived by humans as complex patterns composed of spatially organized, repeated subpatterns, which have characteristic somewhat uniform appearance carries substantial information about the structure of physical objects – analysis is an important issue in image processing and understanding ISITC 2007, Ma. Zda - a software. . . 5

Analysis pathways ISITC 2007, Ma. Zda - a software. . . 6

Image loading Gray-scale images formats to load: – – – Siemens NUMARIS 2 and 3 Siemens Vision ACR NEMA GE Advantage GE IDBM IGE – YMS Bruker Aspect 3000 Picker Dicom Windows Bitmap Unformatted 8 or 16 bits/pixel ISITC 2007, Ma. Zda - a software. . . 7

3 D image loading Loading: – – – Window bitmaps 3 D Bmf format 3 D Dicom data View adjustment: – – – cross-sections selection adjustment of angles zoom gray-scale window gray-scale thresholds ISITC 2007, Ma. Zda - a software. . . 8

Region of interest • • Region of interest (ROI) is a set of pixels in 2 D image or voxels in 3 D image selected for processing. ROIs concentrate computation effort on image fragment that is relevant to a goal of computation and thus helps avoid processing of unnecessary image fragments. 2 D ROI Editor in Ma. Zda a) window title bar, b) menu bar, c) image panel, d) load file button, e) copy and move buttons, f) graphics toolbar for ROI edition, g) morphological tools for ROI edition, h) drawing mode selection buttons, i) ROI color selector, j) ROI on/off switches, k) zoom in/out buttons, l) sliders for adjustment of grey-scale palette, m) image view mode selector, n) status bar ISITC 2007, Ma. Zda - a software. . . 9

3 D ROI Editor Defining ROI with interactive tool of elastic surface ISITC 2007, Ma. Zda - a software. . . ROI found with flood-fill algorithm 10

Textural features computation Options for the analysis Input image (2 D or 3 D) with the defined ROIs (selection of feature groups to compute, algorithms parameters, image normalization options, etc. ) ISITC 2007, Ma. Zda - a software. . . Resulting list of textural features (columns of the report correspond with the defined ROIs) 11

Feature lists analysis A Goal: Finding a way of texture classification ISITC 2007, Ma. Zda - a software. . . Tools: – Combining reports for further analysis – Defining class names of regions (columns) – Selection of most discriminative features 12

Feature selection 1. The number of features computed by Ma. Zda may reach several-hundred per region, which is difficult to handle. 2. The several-hundred features turns into the problem of analysis of a several-hundreddimensional space => statistically not reliable. 3. Usually only a limited number of features carry relevant information needed for texture discrimination. Ma. Zda allows for selection of most effective features and rejection of the others. ISITC 2007, Ma. Zda - a software. . . 13

Feature space visualization selected feature names of classes List of selected features loaded into b 11 module. The features are computed for textures of two different classes. ISITC 2007, Ma. Zda - a software. . . Visualization of feature space in b 11 module. 14

Feature analysis & classification Feature analysis: • Principal Component Analysis • Linear Discriminant Analysis Feature classification • 1 – NN classifier • Artificial neural network (training/testing) • Nonninear Discriminant Analysis ISITC 2007, Ma. Zda - a software. . . 15

Feature Clustering • k-means • Agglomerative Hierarchical Clustering • Similarity Based Clustering ISITC 2007, Ma. Zda - a software. . . Segmentation • k-means 16

Applications • • • brain research - amygdale activation detection research of hipocampal sclerosis investigation of healthy and cirrhotic livers textural analysis of trabecular bone - osteoporosis detection monitoring of atrophy and regeneration of muscles monitoring of teeth implants assessment of cellular necrosis in epithelial cells evaluation of anti-vascular therapy of mammary carcinomas discrimination between cooked and raw potatoes and potato varieties analysis of apple ripening process MRI image analysis of soft cheeses The Ma. Zda package has proven to be an efficient and reliable set of software tools for analysis of textured images. download: http: //www. eletel. p. lodz. pl/merchant/mazda/order 1_en. epl ISITC 2007, Ma. Zda - a software. . . 17