Label‐Free Identification of White Blood Cells Using Machine Learning

Nassar, Mariam; Doan, Minh; Filby, Andrew; Wolkenhauer, Olaf; Fogg, Darin K.; Piasecka, Justyna; Thornton, Catherine A.; Carpenter, Anne E.; Summers, Huw D.; Rees, Paul; Hennig, Holger

doi:10.1002/cyto.a.23794

Cited by 85 publications

(80 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We continued training over the course of one month to promote a model with highest validation accuracy possible, demonstrating stable execution of AID for long run-times. Our results show that the classification performance of AID is at least similar to other publications showing labfel-free discrimination of B-and T-cells (Nassar et al, 2019;Yoon et al, 2018Yoon et al, , 2017. These results suggest feasibility of label-free image-based discrimination of B-and T-cells, which could again be used to complement blood cell counts.…”

Section: Label-free Discrimination Of B-and T-cellssupporting

confidence: 87%

“…Finally, we demonstrate that the tools provided by AID master even challenging classification tasks by training a classifier to distinguish B-and T-cells based on bright-field images from RT-DC. The resulting model reaches a classification performance that is state-of-the-art for label-free approaches (Nassar et al, 2019;Yoon et al, 2018Yoon et al, , 2017. With the demonstrated utility for a wide range of potential image analysis tasks, AID is a ready-to-use software package for anyone who wants to start exploring the power of AI-based image analysis for their own research without the need for any programming skills.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

AIDeveloper: deep learning image classification in life science and beyond

Kräter

Soteriou

Jacobi

et al. 2020

Preprint

View full text Add to dashboard Cite

Publications on artificial intelligence (AI)-based image analysis have increased drastically in recent years. However, all applications use individual solutions highly specialized for a particular task. Here, we present an easy-to-use, adaptable, open source software, called AIDeveloper (AID) to train neural nets (NN) for image classification without the need for programming. The software provides a variety of NN-architectures that can be simply selected for training. AID allows the user to apply trained models on new data, obtain metrics for classification performance, and export final models to different formats. The working principles of AID are first illustrated by training a convolutional neural net (CNN) on a large dataset consisting of images of different objects (CIFAR-10). We further explore the potential of AID by training a model to distinguish areas of differentiated and non-differentiated mesenchymal stem cells (MSCs) in culture. Additionally, we compare a conventional clinical whole blood cell count with a whole blood cell count performed by an NN-trained, using a dataset of more than 1.2 million images obtained by real-time deformability cytometry, delivering comparable results. Finally, we demonstrate how AID can be used for label-free classification of B-and Tcells derived from human blood, which currently requires costly and time-consuming sample preparation. Thus, AID can empower anyone to develop, train, and apply NNs for image classification. Moreover, models can be generated by non-programmers, exported, and used on different devices, which allows for an interdisciplinary use.

show abstract

Section: Label-free Discrimination Of B-and T-cellssupporting

confidence: 87%

Section: Introductionmentioning

confidence: 99%

AIDeveloper: deep learning image classification in life science and beyond

Kräter

Soteriou

Jacobi

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Following the leave-one-donor-out test principle [31,38], we wanted the selection of the optimal hyperparameters to be generalizable to new donors as well. Therefore, we applied a nested cross-validation scheme [55,55,56] (Figure 8).…”

Section: Nested Cross-validationmentioning

confidence: 99%

Classifying T cell activity in autofluorescence intensity images with convolutional neural networks

Wang

Walsh

Skala

et al. 2019

Journal of Biophotonics

View full text Add to dashboard Cite

The importance of T cells in immunotherapy has motivated developing technologies to improve therapeutic efficacy. One objective is assessing antigen‐induced T cell activation because only functionally active T cells are capable of killing the desired targets. Autofluorescence imaging can distinguish T cell activity states in a non‐destructive manner by detecting endogenous changes in metabolic co‐enzymes such as NAD(P)H. However, recognizing robust activity patterns is computationally challenging in the absence of exogenous labels. We demonstrate machine learning methods that can accurately classify T cell activity across human donors from NAD(P)H intensity images. Using 8260 cropped single‐cell images from six donors, we evaluate classifiers ranging from traditional models that use previously‐extracted image features to convolutional neural networks (CNNs) pre‐trained on general non‐biological images. Adapting pre‐trained CNNs for the T cell activity classification task provides substantially better performance than traditional models or a simple CNN trained with the autofluorescence images alone. Visualizing the images with dimension reduction provides intuition into why the CNNs achieve higher accuracy than other approaches. Our image processing and classifier training software is available at https://github.com/gitter-lab/t-cell-classification.

show abstract

“…For instance, to sort T cells in practice, the classifier would need to be coupled to a flow sorter. The flowing nature of the cells may make the resulting images different enough to require a distinct CNN, which is not an obstacle given recent advances in CNNs for imaging flow cytometry 10,21,22,34,48,49 . Overall, our strong results demonstrate the feasibility of classifying T cells directly from autofluorescence intensity images, which can guide future work to bring this technology to pre-clinical and clinical applications.…”

Section: Discussionmentioning

confidence: 99%

“…A trained model must be able to generalize to new donors in order to be useful in a practical pre-clinical or clinical setting. Therefore, all of our evaluation strategies train on images from some donors and evaluate the trained models on separate images from a different donor, which is referred to as subject-wise cross-validation 34 or a leave-one-patient-out scheme 28 . We initially assess the classifiers with cross-validation across donors.…”

Section: /29mentioning

confidence: 99%

Classifying T cell activity in autofluorescence intensity images with convolutional neural networks

Wang

Walsh

Skala

et al. 2019

Preprint

View full text Add to dashboard Cite

The importance of T cells in immunotherapy has motivated developing technologies to better characterize T cells and improve therapeutic efficacy. One specific objective is assessing antigen-induced T cell activation because only functionally active T cells are capable of killing the desired targets. Autofluorescence imaging can distinguish T cell activity states of individual cells in a non-destructive manner by detecting endogenous changes in metabolic co-enzymes such as NAD(P)H. However, recognizing robust patterns of T cell activity is computationally challenging in the absence of exogenous labels or information-rich autofluorescence lifetime measurements. We demonstrate that advanced machine learning can accurately classify T cell activity from NAD(P)H intensity images and that those image-based signatures transfer across human donors. Using a dataset of 8,260 cropped single-cell images from six donors, we meticulously evaluate multiple machine learning models. These range from traditional models that represent images using summary statistics or extract image features with CellProfiler to deep convolutional neural networks (CNNs) pre-trained on general non-biological images. Adapting pre-trained CNNs for the T cell activity classification task provides substantially better performance than traditional models or a simple CNN trained with the autofluorescence images alone. Visualizing the images with dimension reduction provides intuition into why the CNNs achieve higher accuracy than other approaches. However, we observe that fine-tuning all layers of the pre-trained CNN does not provide a classification performance boost commensurate with the additional computational cost. Our software detailing our image processing and model training pipeline is available as Jupyter notebooks at https: //github.com/gitter-lab/t-cell-classification. lifetime imaging to identify macrophages within the tumor microenvironment in vivo 7 and classified activation state of T cells in vitro 6 . The fluorescence lifetime of NAD(P)H and FAD is highly sensitive to the microenvironment and binding of NAD(P)H and FAD. Thus, this fluorescence lifetime can be used to resolve metabolic differences between functional states of immune cells. However, fluorescence lifetime imaging requires specialized and expensive microscope components, limiting its use to particular labs. Although autofluorescence intensity images lack the depth of information provided by the fluorescence lifetime, intensity images can easily and quickly be acquired on almost any commercial fluorescence microscope, allowing widespread adoption and seamless integration of a new technique into existing protocols of live cell assessment. Here, we develop a computational framework that uses autofluorescence intensity images to assess T cell activation state at the single-cell level.Machine learning is promising for classifying cell subtypes from label-free images. For example, Pavillon et al. used regularized logistic regression to predict macrophage activation state 8 , and Yoo...

show abstract

Label‐Free Identification of White Blood Cells Using Machine Learning

Cited by 85 publications

References 14 publications

AIDeveloper: deep learning image classification in life science and beyond

AIDeveloper: deep learning image classification in life science and beyond

Classifying T cell activity in autofluorescence intensity images with convolutional neural networks

Classifying T cell activity in autofluorescence intensity images with convolutional neural networks

Contact Info

Product

Resources

About