Rethinking Cell Counting Methods: Decoupling Counting and Localization

TL;DR

The paper tackles automated cell counting in microscopy by decoupling counting from localization, introducing a Counter that operates on intermediate features to produce a coarse count and density map and a Localizer (UNet) that reconstructs a high-resolution localization map conditioned on the image and coarse map. A Global Message Passing module augments the counting pathway to capture long-range dependencies across the image. Across four datasets, this decoupled scheme with GMP achieves state-of-the-art performance, significantly reducing $MAE$ and $MSE$ compared to prior joint-counting approaches, demonstrating the value of global context and task specialization. The approach is simple, plug-and-play, and requires only standard components, making it practical for broad adoption in microscopy analysis and related counting tasks.

Abstract

Cell counting in microscopy images is vital in medicine and biology but extremely tedious and time-consuming to perform manually. While automated methods have advanced in recent years, state-of-the-art approaches tend to increasingly complex model designs. In this paper, we propose a conceptually simple yet effective decoupled learning scheme for automated cell counting, consisting of separate counter and localizer networks. In contrast to jointly learning counting and density map estimation, we show that decoupling these objectives surprisingly improves results. The counter operates on intermediate feature maps rather than pixel space to leverage global context and produce count estimates, while also generating coarse density maps. The localizer then reconstructs high-resolution density maps that precisely localize individual cells, conditional on the original images and coarse density maps from the counter. Besides, to boost counting accuracy, we further introduce a global message passing module to integrate cross-region patterns. Extensive experiments on four datasets demonstrate that our approach, despite its simplicity, challenges common practice and achieves state-of-the-art performance by significant margins. Our key insight is that decoupled learning alleviates the need to learn counting on high-resolution density maps directly, allowing the model to focus on global features critical for accurate estimates. Code is available at https://github.com/MedAITech/DCL.

Figures (3)

• Figure 1: The pipeline of the proposed method for cell counting. Unlike prior works, we decouple counting and localization, and design an architecture with separate counter and localizer to enable each to specialize on its target task. Moreover, we introduce a global message passing module into the counting network to model long-range spatial dependencies between image regions, enriching feature representations.
• Figure 2: Visualization of the learned single-channel feature map in the counting network. Note that the counter manages to localize cells despite having no cell location annotations at training, just cell count labels.
• Figure 3: Visualization of learned single-channel feature maps from counting networks with and without the proposed global message passing module, along with the estimated cell counts.