Gland segmentation in colorectal cancer histopathological images using U-net inspired convolutional network

The accurate gland segmentation from digitized H&E (hematoxylin and eosin) histology images with a wide range of histologic grades of cancer is quite challenging. The methodologies proposed in recent researches have performed well in segmenting glands from benign subjects but have not given sati...

Full description

Saved in:
Bibliographic Details
Published in:Neural computing & applications Vol. 34; no. 7; pp. 5383 - 5395
Main Authors: Rastogi, Priyanka, Khanna, Kavita, Singh, Vijendra
Format: Journal Article
Language:English
Published: London Springer London 01.04.2022
Springer Nature B.V
Subjects:
Accuracy
Artificial Intelligence
Cancer
Coders
Colorectal cancer
Computational Biology/Bioinformatics
Computational Science and Engineering
Computer Science
Data Mining and Knowledge Discovery
Datasets
Deep learning
Digitization
Encoders-Decoders
Feature maps
Histology
Image Processing and Computer Vision
Image segmentation
Localization
Machine learning
Methods
Metric space
Morphology
Original Article
Probability and Statistics in Computer Science
Semantics
Deep learning
Digital pathology
CNN
Colon cancer
Semantic segmentation
U-net
Gland segmentation
Histology images
Intelligent diagnostics
ISSN:0941-0643, 1433-3058
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The accurate gland segmentation from digitized H&E (hematoxylin and eosin) histology images with a wide range of histologic grades of cancer is quite challenging. The methodologies proposed in recent researches have performed well in segmenting glands from benign subjects but have not given satisfactory results when segmenting glands from malignant cases. The methodology proposed in this paper is based on the symmetric encoder-decoder network which works remarkably well in detecting and segmenting glands in the case of malignant subjects and overall. The proposed pipelines harness the power of multilevel CNN architecture to capture contextual information and concatenation of features from skip connections with upsampled feature maps to improve the localization accuracy thereby improving the precise segmentation accuracy. The raw predicted map is further refined using morphological operators as a post-processing tool. The method is trained and evaluated on the Warwick-QU dataset. The final segmentation results have been compared with the performance of top-performing teams of gland challenge (GlaS) MICCI 2015 along with the recent researches. The final predicted segmentation maps have achieved the F1 score of 0.81 for gland detection, object dice score of 0.82 for segmentation, and Hausdorff distance of 84.18 for gland shape similarity which is akin to or higher than the existing models for the malignant subject. The model has done affably well on the overall dataset too.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0941-0643
1433-3058
DOI:10.1007/s00521-021-06687-z