PuSH - Publication Server of Helmholtz Zentrum München: Classification of normal versus malignant cells in B-ALL microscopic images based on a tiled convolution neural network approach.

Information

Clues to quality of journals

Open Access Policy of Helmholtz Association 2016

CC Licencences

Publication Metrics

Navigation

Home

Deutsch

EVA: Electronic Publishing

New EVA Application

Research

Advanced Search

Browse by ...

... HMGU-Authors/Consortia

... Organizational Structure

... Journal

... Publication Type

... Research Data

... Arbeitsgruppen

... Publication Year

Publication overview

Statistics

Statistics (last 5 years)

OA Publications

Publish

Submit Publication

Import publication from...

...EVA

Report missing publication

Highlights

Support & Contact

Contact persons

Help

Data protection

Helmholtz Open Science

JANE Journal Estimator

SHERPA/RoMEO

DOAJ

Export:

Text

Endnote (RIS) BIB

BibTeX

Mohajerani, P. ; Ntziachristos, V.

Classification of normal versus malignant cells in B-ALL microscopic images based on a tiled convolution neural network approach.

In: Lecture Notes in Bioengineering. 2019. 103-111

DOI

Abstract
Metrics
Extra information

In this paper we present a method based on the existing convolution neural network architecture of AlexNet for the purpose of classifying microscopic images of B-ALL white blood cancer cells. This classification problem is especially challenging due to lack of conspicuous morphological differences between normal and malignant cell nuclei. Therefore, we designed a machine learning pipeline that focused on the texture of the staining images. Briefly, our approach divides the cell image into several overlapping tiles and trains a modified version of AlexNet on the tiles. Only those tiles are retained which are fully contained within the cell image. Several such networks were trained in an ensemble fashion using different training–validation data splits. For a given test image, the tiles are generated and ran through all the trained networks. The outputs of all networks along with the nucleus area are then fed into a simple decision tree, which generates the final prediction. The proposed method was developed in the context of the ISBI 2019 C-NMC challenge. The final testing results demonstrated a classification-weighted F1 score of 0.8307 using 2586 test images. The results demonstrate the possibility of making relatively accurate predictions using only local texture features.

Altmetric