NeuroAdaptTrainer: YOLO-based neuron segmentation plugin for Fiji with transfer learning

Neuron counting and segmentation in microscopy images is a routine yet time-consuming task in neuroscience, often done manually or with semi-automatic tools. To address this, Daniela Eraso-Casas and colleagues developed NeuroAdaptTrainer, an open-source Fiji/ImageJ plugin that embeds a YOLO instance-segmentation model directly into the microscopist's existing workflow. The plugin enables automatic neuron detection on single images or entire batches, eliminating the need to switch between separate deep-learning environments. This integration is particularly valuable because Fiji remains the de facto platform for bioimage analysis, letting researchers stay in a familiar interface while benefiting from state-of-the-art computer vision.

The plugin's key innovation is its interactive correction and transfer learning loop. After running YOLO-based detection, users can manually fix false positives or missed neurons directly in Fiji. These corrections are then used to fine-tune the model to new imaging conditions, adapting it without requiring any coding expertise. An external validation module provides quantitative comparison between the base and adapted models on a held-out annotated dataset, ensuring that updates genuinely improve performance. By combining automated segmentation, expert-in-the-loop correction, and transfer learning, NeuroAdaptTrainer lowers the barrier for non-specialist users to adopt deep learning, while preserving the human oversight essential for reliable biological analysis. The plugin is designed for neuroscientists who need accurate, adaptable neuron segmentation without becoming machine learning experts.