Identyfikatory
Warianty tytułu
Analysis of medical images based on graph search algorithms
Języki publikacji
Abstrakty
W artykule przedstawiono wyniki testów niekonwencjonalnego zastosowania metod do przeszukiwania grafów w celu analizy obrazów powstałych z rezonansu magnetycznego głowy. Zaprezentowano GUI do automatycznej obróbki serii obrazów. Zbudowane klasyfikatory wykazały, że metoda BFS analizy plików DICOM, po odpowiednej selekcji cech, pozwala na 100% rozpoznawanie chorych na wodogłowie i ponad 90% zdrowych, co zachęca do dalszych badań i obserwacji, np. czy osoby sklasyfikowane błędnie jako chorzy, po czasie rzeczywiście nie rozwinęli tej choroby.
There are many methods for image segmentation [1, 2]: threshold, area, edge and hybrid methods. Area methods indicate groups of similar pixels form local regions [3, 4]. Edge methods detect boundaries between homogeneous segments [5, 6, 7]. In this paper we present the results of tests of unconventional implementation of graph search methods for the analysis of images generated from magnetic resonance imaging [8]. We explored the effectiveness of different approaches for dividing areas within a similar gray scale, using adapted graph search algorithms (DFS, BFS) after appropriate modification (Fig. 1). For this purpose, the Weka package (a tool for pre-processing, classification, regression, clustering and data visualization) was used [9]. A training set was generated after analyzing all the series of images from the database. First, we evaluated models created using certain algorithms and compared their efficacy (Tab. 1). This was followed by a selection of attributes (Tab. 2) and a re-evaluation of the models (Tab. 3). Comparison of the results of both evaluations showed that after selection of the relevant product attributes, you can achieve up to 100% detection of patients with hydrocephalus and over 90% proper recognition of healthy persons. This encourages further research and observation, such as whether persons wrongly classified as sick actually developed the disease in time. We designed a web application for the study, written in Windows Azure, as well as a GUI for automatic processing of a series of images (Fig. 2).
Wydawca
Czasopismo
Rocznik
Tom
Strony
578--580
Opis fizyczny
Bibliogr. 9 poz., rys., tab.
Twórcy
Bibliografia
- [1] Ludwiczuk R., Bochniak A.: Metody segmentacji obrazów medycznych; Przetwarzanie informacji w społeczeństwie informacyjnym, S. 95-99 PWSZ, Biała Podlaska, 2007.
- [2] Liu L., Sclaroff S.: Region segmentation via deformable model-guided split and merge. Boston University Computer Science Technical Report, (24), 2000.
- [3] Ding L., Goshtasby A.: On the Canny edge detector, Elsevier, Pattern Recognition, 34, p. 721-725, 2001.
- [4] Wanpeng C., Rensheng C., Dong Y.: An illumination-independent edge detection and fuzzy enhancement algorithm based on wavelet transform for non-uniform weak illumination images, Elsevier, Pattern Recognition, 29, 192–199, 2008.
- [5] Cohen I., Cohen L. D.: Finite-element method for active contour models and balloons for 2D and 3D images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 15:1131-1147, 1993.
- [6] Caselles V., Catte F., Coll T.: A geometric model for active contours in image processing. Numerische Mathematik, 66(1):1-31, 1993.
- [7] Xu C., Prince J. L.: Snakes, shapes, and gradient vector flow. IEEE Transactions on Image Processing, 7(3):359-369, 1998.
- [8] Pianykh O. S.: Digital Imaging and Communications in Medicine (DICOM), Springer-Verlag Berlin Heidelberg, 2008.
- [9] Liu B.: Web Data Mining: Exploring Hyperlinks, Contents, and Usage Data, Springer-Verlag Berlin Heidelberg, 2007.
Typ dokumentu
Bibliografia
Identyfikator YADDA
bwmeta1.element.baztech-article-BSW4-0122-0005