Examination of the application of quantitative analysis of CT brain images in ischaemic stroke and brain tumour detection – preliminary test

1 Klinika Neurologii i Udarów Mózgu, Uniwersytet Medyczny w Łodzi
2 Instytut Elektroniki, Politechnika Łódzka
3 Zakład Diagnostyki i Terapii Radiologicznej i Izotopowej, Szpital Kliniczny im. WAM w Łodzi
Adres do korespondencji: Klinika Neurologii i Udarów Mózgu, Uniwersytet Medyczny w Łodzi, ul. Żeromskiego 113, 90-549 Łódź,
tel./faks: +48 42 639 35 91; Zakład Elektroniki Medycznej, Instytut Elektroniki, Politechnika Łódzka, ul. Wólczańska 211/215,
budynek B9 (II piętro), 90-924 Łódź, tel.: +48 42 631 26 31, 42 631 26 46
Praca finansowana ze środków własnych

AKTUALN NEUROL 2014, 14 (2), p. 89–95
DOI: 10.15557/AN.2014.0009

Introduction: Neuroimaging is a standard examination implemented for diagnosis of various pathologies of the central nervous system. The fundamental diagnostic procedures in medical imaging of the central nervous system are computed tomography and magnetic resonance imaging. In case of a sudden focal or generalized onset of brain dysfunctions at first we should think about stroke. A very important test if stroke is suspected is computed tomography. In this paper we would like to check if it is possible to distinguish two pathologies of the cerebrum: ischaemic stroke and tumour, using quantitative analysis of selected abnormalities. Material and methods: Analysis is based on comparison of two pathologies (ischaemic stroke and tumour). Two sets of images were prepared. Analysis is performed to distinguish abnormalities observed on computed tomography brain images from healthy tissue. The image analysis includes data conversion, normalization of region of interest, estimation of the number of texture features, features selection based on four different methods of selection and finally classification based on artificial neural network classifier. Results: In the examination, different effectiveness of used methods was observed. Quantitative analysis of selected texture features allows to differentiate two classes of pathologies. Also an important observation is that the artificial neural network can be a useful tool in data classification and analysis. Conclusions: The performed analysis is effective but only for small number of data. That is why it still needs to be conducted on a larger set of data. It will be also necessary to repeat classification a number of times and to perform data validation in order to confirm effectiveness of the presented method. After that we can hope to get really satisfying results.

Keywords: ischaemic stroke, tumour, neuroimaging, artificial neural network, multilayer perceptron, computed tomography