This file includes Supplementary Notes 1-5, Supplementary Figs. 1-19, Supplementary Tables 1-4, legends for Supplementary Videos 1-16, and Supplementary References.

Overview of the Airscope. Airscope assembly workflow and wireless neuronal recording. Part 1: A freely moving mouse equipped with Airscope is shown, followed by device assembly, LED illumination, fluorescence collection, rapid battery exchange, battery-powered operation, onboard SD-card storage, and real-time neuronal preview on a computer via Wi-Fi. Part 2: Representative neuronal imaging acquired with Airscope. Scale bar, 500 μm. Part 3: Representative applications, including simultaneous recordings from multiple freely interacting mice in a 5 m² open habitat and dual-site imaging within a single mouse using two Airscopes to record cortical and spinal cord activity simultaneously.

Assessing Airscope’s impact on freely moving behavior. Animals were evaluated in a 400 × 400 × 300 mm3 open-field arena both with a head-bar only (left panels) and while wearing the Airscope (right panels). Top row: Representative raw video frames processed by SAM2Mice for automated animal segmentation. Bottom row: Corresponding motion trajectories extracted from the segmented videos. Scale bar: 10 cm.

Airscope neural recording with high-resolution 3D reconstructed skeletons. Top left: Side-view video of a mouse wearing Airscope, recorded by eight surrounding cameras. Top middle: Overhead behavior video with SAM2Mice segmentation. Top right: 3D skeletons reconstructed from multi-camera views42. Bottom left: Raw and background-removed calcium imaging. Bottom middle: Frame-wise single-neuron connectivity calculated using a 100-frame sliding window, with opacity indicating connection strength. Bottom right: Heatmap of neural activity across brain regions, with example neuronal traces from two regions. Scale bar, 500 μm.

Robust zero-shot segmentation and identity tracking through SAM2Mice. Part I: Zero-shot SAM2Mice segmentation of multiple mice in open-field and food-competition scenarios, compared with ground truth and SAM2. Part II: Prompt-free segmentation using YOLO v11-generated identity prompts in a 2.3 × 2.3 m habitat. Part III: Robust zero-shot performance across diverse behavioral environments. Part IV: Dynamic identity tracking during sequential animal appearance, compared with DeepLabCut. Part V: Multi-animal segmentation for three and five mice, compared with SuperAnimal and DeepLabCut. SAM2Mice maintains robust segmentation and consistent identities under crowded conditions. J&F denotes mean Jaccard index and F-score.

Simultaneous animal group recording with Airscope and multi-animal instance segmentation using SAM2Mice. Top left: Five freely moving mice wearing Airscopes in a 60-cm circular arena, shown as raw video and SAM2Mice segmentation. Dots indicate 2-s centroid trajectories, and highlighted boundaries mark close social interactions (<2 cm). Top middle: Identity-coded segmentation masks with 2-s trajectories. Top right: 3D X–Y–time trajectories derived from mask centroids. Bottom: Simultaneous Airscope recordings from five mice, comprising 21,397 neurons. Heatmaps show cortical neural dynamics with representative calcium traces. Bar colors correspond to animal identities defined by the segmentation masks.

Animal groups wearing Airscope freely exploring large-scale, complex naturalistic environments. Left: Multi-angle videos of a 2.3 × 2.3 m naturalistic habitat containing sandy, floral, and grassland zones, enriched with tree hollows, shelters, staircases, and running wheels. The camera array enables comprehensive behavioral tracking of Airscope-equipped mice from multiple perspectives. Right: Simultaneous Airscope recordings from three freely moving mice. Each panel shows cortex-wide neural activity heatmaps, with enlarged insets displaying representative neuronal calcium traces over time.

Neural recording of social hierarchy competition in tube test through Airscope. Top: Frame-wise manual annotation of tube test social behaviors, capturing the full sequence from the initial still phase to the outcome of the winner, with labeling of four behavioral categories: push, resist, retreat, and still. Middle: Side view of the competitive tube assay. Mice were implanted with Airscope, enabling simultaneous neural recording during social competition. The compact Airscope design permitted unrestricted animal movement within the tube. Bottom: Synchronized calcium signal recordings from both mice. Left panel shows multi-region activity heatmap, with right panels highlighting detailed neuronal traces over time.

Neural recording of object-play and food competition through Airscope. Part I: Object-play competition. Top: Side-view video with SAM2Mice masks of dominant (orange) and subordinate (green) mice, segmented masks, and 10-s movement trajectories; mask boundaries are highlighted during wheel occupancy. Bottom: Multi-region neural activity from both mice, with inset calcium traces, and wheel occupancy quantified by SAM2Mice (GT) and model prediction. Dominant and subordinate occupancy ratios are shown in orange and green, respectively; dashed lines indicate 0.8. Part II: Food competition. Same layout and analysis as in Part I, applied to food competition.

Neural recording of an animal freely exploring a complex maze with Airscope. Top left: Airscope-equipped mouse navigating a 15 × 8 binary maze with SAM2Mice segmentation, showing current position and behavioral state. Top middle: Raw and background-removed calcium imaging. Top right: Movement trajectory with velocity-scaled arrows. Bottom left: Cortical neural activity heatmaps with representative neuronal traces. Bottom middle: Frame-wise single-neuron connectivity calculated using a 100-frame sliding window, with opacity indicating connection strength. Bottom right: Binary-tree representation of maze topology, with the highlighted path dynamically showing the mouse’s sequence of choices. Scale bar, 500 μm.

Neural recording of an animal during forced swimming test (FST) with Airscope. Top left: Manually annotated locomotor behaviors, categorized into floating, struggling, and swimming (Methods). The white line in the video represents the current time. Bottom left: A mouse equipped with an Airscope performing forced swim test in a cylindrical chamber with a diameter of 20 cm (left). The corresponding segmented mask video generated by SAM2Mice is exhibited (right). Right: Heatmap of neural activity across different brain regions during FST, with traces sampled from two cortical regions presented on the right.

Neural recording of an animal during tail suspension test (TST) with Airscope. Top left: Raw and SAM2Mice-masked videos during TST. Top right: X- and Y-axis motion traces, with pink indicating active struggling and black indicating passive periods; the red line marks current time. Bottom left: Multi-region neural activity heatmaps with representative neuronal traces. Bottom right: Mean activity of eight selective neurons over time, showing elevated calcium activity during active struggling and reduced activity during passive periods.

Neural recording of an animal during head rotation test with Airscope. Top left and middle: A mouse with Airscope was rotated back and forth at 120° step. Top right: The angular velocities (9.6°/s, 50.4°/s, 91.2°/s) of the rotation test setup. Bottom left: Background removed cortical recordings from Airscope. Bottom middle: Heatmap of neural activity across different brain regions, with traces sampled from two cortical regions presented on the right. Bottom right: Averaged dorsal cortex-wide neuronal activity from n = 8 selective neurons. Pink region indicates animal rotation.

Simultaneous cortex and spinal cord vascular imaging through Airscope. Mice equipped with dual Airscopes for concurrent capture of cortical and spinal cord vasculature. Left panels: overhead video (top) and locomotor trajectories. Middle panels: cortical vascular network (top) with corresponding vessel dilation traces (bottom). Right panels: spinal cord vascular network (top) and associated dilation traces (bottom). Scale bar: 500 μm in main panels and 200 μm in zoom-in panels.

Simultaneous cortex and spinal cord neural imaging through Airscope. Mice wearing dual Airscopes for simultaneous recording of cortical neuronal and spinal cord population activity. Left panels: overhead video (top) and locomotor trajectories. Middle panels: cortical calcium activity maps (top) and spinal cord activity video frames (bottom). Right panels: extracted neuronal populations across distinct cortical regions (top) and dynamics of spinal cord functional components (bottom). Scale bar: 500 μm.

Integrated recording of neural activity, locomotion, and orofacial dynamics in freely moving mice through Airscope. Top left: An Airscope-equipped mouse exploring a 60 × 60 cm² open field, shown with raw video and SAM2Mice segmentation. Top middle: Orofacial video with DeepLabCut-tracked pupil, eye, and nose keypoints. Top right: Pupil position and area, with blinks indicated in red. Bottom left: Multi-region neural activity heatmaps with representative traces. Bottom middle: Raw and background-removed calcium imaging. Bottom right: Nose position and locomotion speed, with self-grooming periods indicated in blue. Scale bar, 500 μm.

Simultaneous neural and first-person view recording of a freely moving animal with Airscope. Top left: Egocentric visual input recorded by an Airscope-modified head-mounted camera during arena exploration. Top middle: Airscope-equipped mice exploring a 60-cm circular arena with four visual wall cues, shown as raw video and SAM2Mice segmentation. Top right: CEBRA embedding of neural activity into 3D latent space, color-coded by visual cue. Bottom left: Raw and background-removed calcium imaging. Bottom middle: Multi-region neural activity heatmaps with representative traces. Bottom right: Dynamic CEBRA latent trajectories over time, with colors indicating the corresponding visual cues. Scale bar, 500 μm.