DETECTION AND CLASSIFICATION OF LEUKOCYTE CELLS FROM SMEAR IMAGE


Kasim O., Kuzucuoglu A. E.

JOURNAL OF THE FACULTY OF ENGINEERING AND ARCHITECTURE OF GAZI UNIVERSITY, cilt.30, sa.1, ss.95-109, 2015 (SCI İndekslerine Giren Dergi) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 30 Konu: 1
  • Basım Tarihi: 2015
  • Dergi Adı: JOURNAL OF THE FACULTY OF ENGINEERING AND ARCHITECTURE OF GAZI UNIVERSITY
  • Sayfa Sayıları: ss.95-109

Özet

Diagnosis of diseases which are related to disorders of the leukocyte's structure is based on morphological examination methods. At the end of this examination, leukocyte's numbers, disorders and types of them are determined by hematology specialists without an intense pace for staining and lighting effort. Therefore, this process causes the loss of the vital information and time. The algorithm which is proposed in this study can support the specialists to examine the not well stained smear easily. Thus, specialists and proposed algorithm can observe the cells' structure on a clear image and classify them into five categories without loss of information. In this manner, quick examination, loss of time, inadequate specialists and diagnosis problems can be solved by the proposed algorithm. At first step of our algorithm, the leukocyte's area is enriched with Otsu based dynamic Piecewise Linear Filtering Method. After that a hybrid spatial learning structure which is composed of K-Means, Markov Random Field and Maximization Expectation Method has been used to get Region of Interests. This hybrid method minimizes the quality of different staining and lighting problems. At analysis step, we obtained 34 different vector elements of each Region of Interests. The member of this vector is dropped to 11 by Gini Method. The Cnt Factor and Sc Factor which are proposed in this study decrease this number of feature set into 5. This also reduces the classification pace. Then, this vector is divided into 5 different classes with Probabilistic Neural Network. Classification performance of the allocated data set is measured as 91.66%. The obtained results can support the specialist and the proposed algorithm can give useful information about the smear if there isn't any specialist.