THE USE OF CLASSIFICATION TREES FOR OBJECT RECOGNITION ON DIGITAL IMAGES OF MICROSCOPIC PREPARATIONS

2018

В мире научных открытий, Vol 10, Iss 3, Pp 12-23 (2018)

Background. The construction of mathematical models of classification trees for object recognition in digital microscopic images of sputum stained by the method of Ziehl-Nielsen. Materials and methods. Data on 177,393 objects isolated on digital images of microscopic preparations were used: 6,708 objects – acid-fast mycobacteria, 170,685 – other objects. The analysis of objects was carried out on 240 color and morphometric features. Classification of objects used classification trees, built by different methods. Result. The highest degree of accuracy is possessed by the classification tree constructed by the method of Comprehensive CHAID, but this tree has a lower sensitivity index than the classification tree constructed by the CHAID method. In addition, the latter classification tree includes a smaller number of object parameters required for classification. The sensitivity of the classification tree constructed by the CHAID method was 94.0 [93.4; 94.6]%, specificity 92.1 [92.0; 92.1]%, accuracy – 92.2 [92.1; 92.3]%. Conclusion. Constructed using different methods, classification trees allow automatic recognition of objects isolated on digital microscopic images of sputum stained using the Ziehl-Nielsen method. At the same time, the classification tree constructed by the CHAID method has the best indicators characterizing the diagnostic ability of these models.

article

English
Russian

Science and Innovation Center Publishing House, 2018.

деревья классификации, математическое моделирование, кислотоустойчивые микобактерии, бактериоскопическая диагностика, Agriculture, Science