ISSN 2079-3537      

 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                                                                                             

Scientific Visualization, 2019, volume 11, number 3, pages 64 - 75, DOI: 10.26583/sv.11.3.06

Trainable Active Contour Model for Histological Image Segmentation

Authors: A.V. Khvostikov1,A, A.S. Krylov2,A, I.A. Mikhailov3,B, P.G. Malkov4,B

A Laboratory of Mathematical Methods of Image Processing, Faculty of Computational Mathematics and Cybernetics, Lomonosov Moscow State University

B Department of Pathology, University Medical Center, Lomonosov Moscow State University

1 ORCID: 0000-0002-4217-7141, khvostikov@cs.msu.ru

2 ORCID: 0000-0001-9910-4501, kryl@cs.msu.ru

3 ORCID: 0000-0001-8020-369X, imihailov@mc.msu.ru

4 ORCID: 0000-0001-5074-3513, pmalkov@mc.msu.ru

 

Abstract

Lesions analysis of mucous glands, which depends on the glands segmentation in histological images, is an important task of surgical pathology. This paper presents a hybrid method of glands object segmentation in histological images, based on the trainable active contour model. The hybrid method combines the use of both modern convolutional neural networks and classical methods of mathematical image processing. Also, within this hybrid method a special postprocessing algorithm is implemented, which allows to correctly segment stucked glands in the image. The proposed method was tested on PATH-DT-MSU dataset and demonstrated good results. The average value of IoU for all test images is 0.81.

 

Keywords: image segmentation, hybrid algorithms, convolutional neural networks, active contours, histological images.