USING CONVOLUTION NEURAL NETWORK FOR CLASSIFICATION OF DIFFERENT TISSUE IMAGES IN HISTOLOGICAL SECTIONS

Nowadays, medical images have been studied extensively using artificial intelligence methods. Especially deep learning models have been used frequently in diagnostic and identification systems in medical images. Many researchers and health experts work on the identification, counting and morphologic...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Annals of Faculty Engineering Hunedoara 2019-02, Vol.17 (1), p.101-104
Hauptverfasser: Inik, Özkan, Balcioğlu, Esra, Ceyhan, Ayşe, Ülker, Erkan
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Nowadays, medical images have been studied extensively using artificial intelligence methods. Especially deep learning models have been used frequently in diagnostic and identification systems in medical images. Many researchers and health experts work on the identification, counting and morphological characterization of tissue cells in histological sections. When these tasks are performed manually, time and workload are increased. For this reason, a fully automatic system is required to be made on the cells in the tissues. For a comprehensive operation of the system to be designed, it is first necessary to identify the tissues and then work on the cells forming the tissue. Due to this reason, in this study 5 different tissues were automatically classified. These tissues are Eye, Kidney, Liver, Ovary and Cerebellum respectively. A new Convolution Neural Network (CNN) is designed for classification. The input image size of the network is 220x220x3 and consists of 24 layers in total. A new data set has been created for this CNN training and testing. This data set is composed of 20500 images each class and totally 102500 image. 60% of these images were used for training and the remaining were used for testing. As a result of the study, it was seen that 5 different tissue were estimated with 96.47% accuracy with designed CNN.
ISSN:1584-2665
2601-2332