CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
Modeling morphological structures on radiography of thorax in intelligent diagnostic systems for medical purposes
 |
Read | Kudryavtsev P.S., Kuzmin A.A., Savinov D.Yu., Filist S.A., Shatalova O.V. Modeling morphological structures on radiography of thorax in intelligent diagnostic systems for medical purposes // Caspian journal : management and high technologies. — 2017. — №3. — pp. 109-120. |
/109-120.jpg)
Kudryavtsev P.S. - postgraduate, Southwest State University, 19B Chelyuskintsev St., Kursk, 305004, Russian Federation, 79pavel97@mail.ru
Kuzmin A.A. - Cand. Sci. (Engneering), Associate Professor, Southwest State University, 19B Chelyuskintsev St., Kursk, 305004, Russian Federation, ku3bmin@gmail.com
Savinov D.Yu. - postgraduate, Southwest State University, 19B Chelyuskintsev St., Kursk, 305004, Russian Federation, marina-savinova-93@mail.ru
Filist S.A. - Doct. Sci. (Engineering), Professor, Southwest State University, 19B Chelyuskintsev St., Kursk, 305004, Russian Federation, SFilist@gmail.com
Shatalova O.V. - Cand. Sci. (Engneering), Associate Professor, Southwest State University, 19B Chelyuskintsev St., Kursk, 305004, Russian Federation, shatРѕlg@mail.ru
Differential diagnosis of cancer and pneumonia with the help of the image on radiographs of thorax (IRT) is a challenging task. To solve this problem, it requires representative training samples obtained on IRT patients with these diseases, which are then used in the IRРў classifiers. For this one needs to select IRРў on the type of morphological formations (MF) with a certain disposition or with concomitant pathologies. It's a very complicated and time-consuming process. Therefore, it was proposed to simulate the morphological formation necessary for formation of the training samples for configuring neural networks for classification of x-ray images. According to the proposed method of construction of models of MF, statistical analysis of the Walsh spectra in multiscale Windows was carried out. The idea of forming the model of the MF associated with nosology , is the following. On the current IRРў image allocated area (rectangular) L1ВґL2 , which generate a model of the MF of the specified class. Then is by "fit" of the spectrum of each window M , formed around the current pixel to the reference window. Given the adopted structure of the window, M : 16С…16; 32С…32 and 64С…64 pixel around each pixel of the IRРў, are formed a window of three types. Filling region L1ВґL2 pixels corresponding to the selected model, being from the big by size window M . Define the spectral coefficient in this window for the current pixel m window L1ВґL2 and we minimize the Euclidean distance between the current spectrum and etalon. Upon reaching a satisfactory "fit" of the spectrums, switch to the window k= 2 and also optimize the spectral ratio. After this, proceed to k= 1.This procedure can be performed in a loop until a functional characterizing the quality of the "fit" will not be acceptable for all k . A method of modeling morphological structures on chest radiography has been proposed in the result of the research. Method allows one to create training data for the classifiers of x-rays for a given pathology.
Key words: рентгеновский снимок, модель морфологического образования, спектр Уолша, пиксель, классификация изображения, окно анализа, алгоритм построения модели, показатели качества сегментации изображений, x-ray, model of morphological formation, Walsh spectrum, th