Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Single Image Estimation of Cell Migration Direction by Deep Circular Regression

Lennart Bruns, Lucas Lamparter, Milos Galic, Xiaoyi Jiang

TL;DR

This work tackles estimating cell migration direction from a single image (SIECMD) using deep circular regression to respect angular data. It systematically evaluates encoding, activation, and loss configurations with a probing CNN, selects circle encoding with activation $\varphi_{sigmoid2d}$ and loss $\delta_{dist}^2$, and then fine-tunes EfficientNetV2 with ImageNet pretraining, incorporating test-time augmentation. On NIH3T3, U373, and MS3T3 datasets, the method achieves mean angular deviations of approximately $17^\circ$ on NIH3T3 and U373 and around $30^\circ$ on MS3T3, significantly outperforming Nishimoto2019's $30^\circ$–$34^\circ$ results. The work enables high-throughput, tracking-free analyses in cell migration and points to future extensions to more datasets and real-world workflows.

Abstract

In this paper, we address the problem of estimating the migration direction of cells based on a single image. A solution to this problem lays the foundation for a variety of applications that were previously not possible. To our knowledge, there is only one related work that employs a classification CNN with four classes (quadrants). However, this approach does not allow for detailed directional resolution. We tackle the single image estimation problem using deep circular regression, with a particular focus on cycle-sensitive methods. On two common datasets, we achieve a mean estimation error of $\sim\!17^\circ$, representing a significant improvement over previous work, which reported estimation error of $30^\circ$ and $34^\circ$, respectively.

Single Image Estimation of Cell Migration Direction by Deep Circular Regression

TL;DR

This work tackles estimating cell migration direction from a single image (SIECMD) using deep circular regression to respect angular data. It systematically evaluates encoding, activation, and loss configurations with a probing CNN, selects circle encoding with activation and loss , and then fine-tunes EfficientNetV2 with ImageNet pretraining, incorporating test-time augmentation. On NIH3T3, U373, and MS3T3 datasets, the method achieves mean angular deviations of approximately on NIH3T3 and U373 and around on MS3T3, significantly outperforming Nishimoto2019's – results. The work enables high-throughput, tracking-free analyses in cell migration and points to future extensions to more datasets and real-world workflows.

Abstract

In this paper, we address the problem of estimating the migration direction of cells based on a single image. A solution to this problem lays the foundation for a variety of applications that were previously not possible. To our knowledge, there is only one related work that employs a classification CNN with four classes (quadrants). However, this approach does not allow for detailed directional resolution. We tackle the single image estimation problem using deep circular regression, with a particular focus on cycle-sensitive methods. On two common datasets, we achieve a mean estimation error of , representing a significant improvement over previous work, which reported estimation error of and , respectively.
Paper Structure (5 sections, 7 equations, 2 figures, 5 tables)

This paper contains 5 sections, 7 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. SIECMD problem
  3. Method
  4. Experimental validation
  5. Conclusion

Figures (2)

  • Figure 1: Example cell images with ground truth cell migration direction (red arrow). Top: NIH3T3 dataset. Middle: U373 dataset. Bottom: MS3T3 dataset.
  • Figure 2: TTA for migration direction estimation ($n=10$).