NeuroSeg-MF: robust neuron segmentation in two-photon Ca2+ imaging using multi-feature fusion and detection-guided SAM.

Two-photon Ca2+ imaging enables large-scale recording of neuronal activity in vivo, yet reliable neuron segmentation remains challenging in recordings with low contrast, densely packed neurons, and weak activity. Here, we present NeuroSeg-MF (Neuron Segmentation with Multi-feature Fusion), a framework that combines multi-feature fusion with the prompt-based segment anything model (SAM). NeuroSeg-MF integrates multiple spatiotemporal features, including average projection images, pseudo-depth maps, and correlation maps, to facilitate precise neuron detection and subsequent SAM-based segmentation. Experimental results demonstrate that multi-feature fusion enhances detection accuracy, while detection-guided SAM ensures precise neuron segmentation. The proposed framework achieves robust performance across multiple two-photon Ca2+ imaging datasets, thereby providing a practical solution for analyzing data under challenging imaging conditions.