Mitosis Detection Using Convolutional Neural Network Based Features

Albayrak A., BİLGİN G.

17th IEEE International Symposium on Computational Intelligence and Informatics (CINTI), Budapest, Hungary, 17 - 19 November 2016, pp.335-339 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/cinti.2016.7846429
  • City: Budapest
  • Country: Hungary
  • Page Numbers: pp.335-339
  • Yıldız Technical University Affiliated: Yes


Breast cancer is the second leading cause of cancer death in women according to World Health Organization (WHO). Development of computer aided diagnostic (CAD) systems has great importance as a secondary reader systems for a correct diagnosis and treatment process. In this paper, a deep learning based feature extraction method by convolutional neural network (CNN) is proposed for automated mitosis detection for cancer diagnosis and grading by histopathological images. The proposed framework is tested on the MITOS data set provided for a contest on mitosis detection in breast cancer histological images released for research purposes in International Conference on Pattern Recognition (ICPR' 2014). By using provided histopathological images, cellular structures are initially found by combined clustering based segmentation and blob analysis after preprocessing step. Then, obtained cellular image patches are cropped automatically from the histopathological images for feature extraction stage. CNN, which is a prominent deep learning method on image processing tasks, is utilized for extracting discriminative features. Due to the high dimensional output of the CNN, combination of PCA and LDA dimension reduction methods are performed respectively for regularization and dimension reduction process. Afterwards, a robust kernel based classifier, support vector machine (SVM), is used for final classification of mitotic and non-mitotic cells. The test results on MITOS data set prove that the proposed framework achieved promising results for mitosis detection on histopathological images.