CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
AUTOMATED SYSTEM FOR CLASSIFYING BREAST DISEASES BY X-RAY MAMMOGRAPHY
 |
Read | Dabagov Anatoly R., Malyutina Irina A., Kondrashov Dmitry S., Serebrovsky Vadim V., Chahine Mohamad T. AUTOMATED SYSTEM FOR CLASSIFYING BREAST DISEASES BY X-RAY MAMMOGRAPHY // Caspian journal : management and high technologies. — 2019. — №4. — pp. 10-24. |
/10-24.jpg)
Dabagov Anatoly R. - Cand. Sci. (Engineering), President of JSC “Medical Technologies Ltd”, JSC “Medical Technologies Ltd”, dar@mti.ru
Malyutina Irina A. - post-graduate student, Southwest State University, 94 50 let Oktyabrya St., Kursk, 305040, Russian Federation, Irina92_2010@mail.ru
Kondrashov Dmitry S. - post-graduate student, Southwest State University, 94 50 let Oktyabrya St., Kursk, 305040, Russian Federation, kondrashov012@mail.ru
Serebrovsky Vadim V. - Doct. Sci. (Engineering), Professor, Southwest State University, 94 50 let Oktyabrya St., Kursk, 305040, Russian Federation, SFilist@mail.ru
Chahine Mohamad T. - Cand. Sci. (Engineering), Associate Professor, Kursk State Medical University, 3 Karl Marks St., Kursk, 305041, Russian Federation, chahine@Kgmu.com
The aim of the study is to develop classifiers for an automated system of screening classification of mammograms of the breast .A three-stage classification of breast x-rays was investigated. At the first stage, semantic segmentation of the image is carried out. At the second stage, an area of interest is formed from the selected segments, and at the third stage, a decision is made on whether the area of interest belongs to a known morphological formation. The LPR can intervene in the classification process, both in the second and third stages. The automated system uses a two-alternative classifier based on the maximum likelihood method. The classifier includes an alphabet of two classes C1 and C2. Class C1 corresponds to blocks in the normal state, and class C2-these are blocks that have morphological formations caused by pathological processes. Test mammographic images of the breast from the MIAS database with confirmed diagnoses were used as input data for the calculation of informative indicators. Experimental approbation of the software of the automated system on classification of radiographs of a mammary gland on classes "there is no area of interest" or "there is an area of interest"is carried out. As textural characteristics of the classified segments, it is proposed to use statistical characteristics of the brightness of the pixels of the segment: mode, expectation, standard deviation. The exploratory analysis of textural characteristics of samples of two classes: C1 - norm, and C2 - neoplasm is carried out. A criterion for evaluating the results of classification of mammographic images was obtained. Experiments on control samples showed diagnostic efficiency for classes of radiographs "there is no area of interest" - "there is an area of interest" not lower than 90 %, and for classes of segments "norm" - "pathology" not lower than 91 %.
Key words: онкологические заболевания, рентгенограммы молочной железы, сегментация, морфологические образования, область интереса, классификатор, метод максимального правдоподобия, показатели качества классификации, cancers, breast X-rays, segmentation, morphologica