Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On the importance of local and global feature learning for automated measurable residual disease detection in flow cytometry data

Lisa Weijler, Michael Reiter, Pedro Hermosilla, Margarita Maurer-Granofszky, Michael Dworzak

Abstract

This paper evaluates various deep learning methods for measurable residual disease (MRD) detection in flow cytometry (FCM) data, addressing questions regarding the benefits of modeling long-range dependencies, methods of obtaining global information, and the importance of learning local features. Based on our findings, we propose two adaptations to the current state-of-the-art (SOTA) model. Our contributions include an enhanced SOTA model, demonstrating superior performance on publicly available datasets and improved generalization across laboratories, as well as valuable insights for the FCM community, guiding future DL architecture designs for FCM data analysis. The code is available at \url{https://github.com/lisaweijler/flowNetworks}.

On the importance of local and global feature learning for automated measurable residual disease detection in flow cytometry data

Abstract

This paper evaluates various deep learning methods for measurable residual disease (MRD) detection in flow cytometry (FCM) data, addressing questions regarding the benefits of modeling long-range dependencies, methods of obtaining global information, and the importance of learning local features. Based on our findings, we propose two adaptations to the current state-of-the-art (SOTA) model. Our contributions include an enhanced SOTA model, demonstrating superior performance on publicly available datasets and improved generalization across laboratories, as well as valuable insights for the FCM community, guiding future DL architecture designs for FCM data analysis. The code is available at \url{https://github.com/lisaweijler/flowNetworks}.

Paper Structure

This paper contains 35 sections, 1 equation, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Preleminaries
  5. Methods assessed
  6. No-context model
  7. MLP:
  8. Global-context models
  9. MLP-mean:
  10. MLP-max:
  11. MLP-pma:
  12. PointNet:
  13. PointNet-adapted:
  14. ST:
  15. reluFormer:
  16. ...and 20 more sections

Figures (3)

  • Figure 1: This figure shows 2D projections of an FCM sample on pairs of features, where each dot represents the feature vector of a cell (event). Healthy cells are denoted in grey, and cancerous cells in red. FSC, SSC stands for forward-, side-scatter, and CD for cluster of differentiation.
  • Figure 2: This figure shows the general architecture of the proposed local- and global-context model. For our experiments, we used one GNN layer and three ISAB with FPS instead of learned query vectors. The prediction head MLP is a linear layer.
  • Figure 3: Network features of the last layer of each model before the prediction head (a linear layer) are plotted using PCA with two components. Each model had the same FCM sample as input. Healthy cells are denoted in grey, and cancerous cells in red.

Theorems & Definitions (1)

  • definition 1