Evaluation of deep learning training strategies for the classification of bone marrow cell images

Glüge, Stefan; Balabanov, Stefan; Koelzer, Viktor Hendrik; Ott, Thomas

doi:10.1016/j.cmpb.2023.107924

Please use this identifier to cite or link to this item: https://doi.org/10.21256/zhaw-29212

Publication type:	Article in scientific journal
Type of review:	Peer review (publication)
Title:	Evaluation of deep learning training strategies for the classification of bone marrow cell images
Authors:	Glüge, Stefan Balabanov, Stefan Koelzer, Viktor Hendrik Ott, Thomas
et. al:	No
DOI:	10.1016/j.cmpb.2023.107924 10.21256/zhaw-29212
Published in:	Computer Methods and Programs in Biomedicine
Issue:	243
Page(s):	107924
Issue Date:	2023
Publisher / Ed. Institution:	Elsevier
ISSN:	0169-2607 1872-7565
Language:	English
Subjects:	Hematopoiesis; In-domain pre-training; Deep learning; Hematopathology
Subject (DDC):	006: Special computer methods 610.28: Biomedicine, biomedical engineering
Abstract:	Background and Objective: The classification of bone marrow (BM) cells by light mi- croscopy is an important cornerstone of hematological diagnosis, performed thousands of times a day by highly trained specialists in laboratories worldwide. As the manual evaluation of blood or BM smears is very time-consuming and prone to inter-observer variation, new reliable automated systems are needed. Methods: We aim to improve the automatic classification performance of hematolog- ical cell types. Therefore, we evaluate four state-of-the-art Convolutional Neural Net- work (CNN) architectures on a dataset of 171, 374 microscopic cytological single-cell images obtained from BM smears from 945 patients diagnosed with a variety of hema- tological diseases. We further evaluate the effect of an in-domain vs. out-of-domain pre-training, and assess whether class activation maps provide human-interpretable ex- planations for the models’ predictions. Results: The best performing pre-trained model (Regnet y 32gf) yields a mean pre- cision, recall, and F1 scores of 0.787 ± 0.060, 0.755 ± 0.061, and 0.762 ± 0.050, re- spectively. This is a 53.5% improvement in precision and 7.3% improvement in recall over previous results with CNNs (ResNeXt-50) that were trained from scratch. The out-of-domain pre-training apparently yields general feature extractors/filters that ap- ply very well to the BM cell classification use case. The class activation maps on cell types with characteristic morphological features were found to be consistent with the explanations of a human domain expert. For example, the Auer rods in the cytoplasm were the predictive cellular feature for correctly classified images of faggot cells. Conclusions: Our study provides data that can help hematology laboratories to choose the optimal training strategy for blood cell classification deep learning mod- els to improve computer-assisted blood and bone marrow cell identification. It also highlights the need for more specific training data, i.e. images of difficult-to-classify classes, including cells labeled with disease information.
URI:	https://digitalcollection.zhaw.ch/handle/11475/29212
Fulltext version:	Published version
License (according to publishing contract):	CC BY 4.0: Attribution 4.0 International
Departement:	Life Sciences and Facility Management
Organisational Unit:	Institute of Computational Life Sciences (ICLS)
Appears in collections:	Publikationen Life Sciences und Facility Management

Files in This Item:

File	Description	Size	Format
2023_Gluege-etal_Deep-learning-training-strategies-classification-bone-marrow-cell-images.pdf	Article	3.02 MB	Adobe PDF	View/Open
2023_Gluege-etal_Blood-cell-classification_Suppl.pdf	Supplementary material	4.41 MB	Adobe PDF	View/Open

Show full item record

Glüge, S., Balabanov, S., Koelzer, V. H., & Ott, T. (2023). Evaluation of deep learning training strategies for the classification of bone marrow cell images. Computer Methods and Programs in Biomedicine, 243, 107924. https://doi.org/10.1016/j.cmpb.2023.107924

Glüge, S. et al. (2023) ‘Evaluation of deep learning training strategies for the classification of bone marrow cell images’, Computer Methods and Programs in Biomedicine, (243), p. 107924. Available at: https://doi.org/10.1016/j.cmpb.2023.107924.

S. Glüge, S. Balabanov, V. H. Koelzer, and T. Ott, “Evaluation of deep learning training strategies for the classification of bone marrow cell images,” Computer Methods and Programs in Biomedicine, no. 243, p. 107924, 2023, doi: 10.1016/j.cmpb.2023.107924.

GLÜGE, Stefan, Stefan BALABANOV, Viktor Hendrik KOELZER und Thomas OTT, 2023. Evaluation of deep learning training strategies for the classification of bone marrow cell images. Computer Methods and Programs in Biomedicine. 2023. Nr. 243, S. 107924. DOI 10.1016/j.cmpb.2023.107924

Glüge, Stefan, Stefan Balabanov, Viktor Hendrik Koelzer, and Thomas Ott. 2023. “Evaluation of Deep Learning Training Strategies for the Classification of Bone Marrow Cell Images.” Computer Methods and Programs in Biomedicine, no. 243: 107924. https://doi.org/10.1016/j.cmpb.2023.107924.

Glüge, Stefan, et al. “Evaluation of Deep Learning Training Strategies for the Classification of Bone Marrow Cell Images.” Computer Methods and Programs in Biomedicine, no. 243, 2023, p. 107924, https://doi.org/10.1016/j.cmpb.2023.107924.