Patch-Based Encoder-Decoder Architecture for Automatic Transmitted Light to Fluorescence Imaging Transition: Contribution to the LightMyCells Challenge

TL;DR

The paper addresses automatic fluorescence prediction from label-free transmitted-light images using a patch-based RUNet encoder-decoder framework. It conducts thorough ablations across architecture, training strategy, and loss functions, showing that separate per-organelle models with a 512×512 patch size and a multi-term loss yield competitive results on the LightMyCells benchmark. The approach is complemented by open-source code and inference containers, underscoring reproducibility and practical deployment. These findings highlight that carefully chosen data handling and objective composition can match top methods despite data heterogeneity and sparse annotations, with potential for metadata-driven improvements in future work.

Abstract

Automatic prediction of fluorescently labeled organelles from label-free transmitted light input images is an important, yet difficult task. The traditional way to obtain fluorescence images is related to performing biochemical labeling which is time-consuming and costly. Therefore, an automatic algorithm to perform the task based on the label-free transmitted light microscopy could be strongly beneficial. The importance of the task motivated researchers from the France-BioImaging to organize the LightMyCells challenge where the goal is to propose an algorithm that automatically predicts the fluorescently labeled nucleus, mitochondria, tubulin, and actin, based on the input consisting of bright field, phase contrast, or differential interference contrast microscopic images. In this work, we present the contribution of the AGHSSO team based on a carefully prepared and trained encoder-decoder deep neural network that achieves a considerable score in the challenge, being placed among the best-performing teams.

Figures (1)

• Figure 1: Visualization of the inference and training pipeline..