Determining the scale of image patches using a deep learning approach

Detecting the scale of histopathology images is important because it allows to exploit various sources of information to train deep learning (DL) models to recognise biological structures of interest. Large open access databases with images exist, such as The Cancer Genome Atlas (TCGA) and PubMed Ce...

Full description

Saved in:
Bibliographic Details
Main Authors: Otalora, Sebastian, Perdomo, Oscar, Atzori, Manfredo, Andersson, Mats, Jacobsson, Ludwig, Hedlund, Martin, Muller, Henning
Format: Conference Proceeding
Language:English
Subjects:
Biological system modeling
Biomedical imaging
Biopsy
Cancer
Computer architecture
Deep Learning
DenseNet
Digital Pathology
Machine learning
Magnification
Prostate
Scale
Training
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Detecting the scale of histopathology images is important because it allows to exploit various sources of information to train deep learning (DL) models to recognise biological structures of interest. Large open access databases with images exist, such as The Cancer Genome Atlas (TCGA) and PubMed Central but very few models can use such datasets because of the variability of the data in color and scale and a lack of metadata. In this article, we present and compare two deep learning architectures, to detect the scale of histopathology image patches. The approach is evaluated on a patch dataset from whole slide images of the prostate, obtaining a Cohen's kappa coefficient of 0.9897 in the classification of patches with a scale of 5×, 10× and 20×. The good results represent a first step towards magnification detection in histopathology images that can help to solve the problem on more heterogeneous data sources.
ISSN:1945-8452
DOI:10.1109/ISBI.2018.8363703