You are here: Home Research flowCat

FlowCat

Automated classification of B-cell lymphoma sub types

In our project, we seek to establish an approach for automated classification of lymphoma subtypes through a deep-learning based predictive model using information from flow cytometry data thereby reducing the need for manual gating.

The framework uses a deep neural network trained on Multi-channel flow cytometry (MFC) data obtained from more than 25k patients with 9 different B-cell lymphoma subtypes.

Compensated and transformed flow cytometry data from a single case is used to train a self-organizing map (SOM), which allows for dimensionality reduction by assigning rows with similar features to the same node on a grid of nodes. The generated SOM is then used as input to a convolutional neural network generating class predictions. The model is trained from scratch, with 10% of our initial dataset used for validation.

The performance of the classifier is assessed by computing the weighted F1 score. Our approach currently achieves overall weighted F1-scores of 0.92 in an 8-class classification.

 

workflow

 

The latest version of this classifier is always available as a service from res mechanica. If you are interested in running FlowCat in your lab, please contact [Email protection active, please enable JavaScript.] and [Email protection active, please enable JavaScript.] with subject "FlowCat"

 

Collaborators: MLL (Münchner Leukämielabor GmbH), res mechanica

Document Actions