Summary: While people naturally “read” emotions on faces, scientists have struggled to quantify how subtle facial muscle movements map to internal brain states. Researchers have now introduced Cheese3D, an AI-powered discovery platform that combines high-speed cameras and machine learning to track minute facial expressions in mice with precision comparable to EEG measurements — without touching the animal.
Cheese3D blends advanced computer vision, synchronized multi-camera imaging, and interpretable machine learning to convert rapid 2D footage into precise 3D facial motion data. This enables researchers to study facial expressions, behavior, and underlying neural states in mice with unprecedented spatial and temporal resolution.
Key Facts
- The Cheese3D Rig: A calibrated array of six synchronized mini-cameras films a mouse’s face from multiple angles, overcoming the difficulty of the mouse’s small, cone-shaped anatomy.
- AI Synthesis: Machine learning models assemble the multi-view footage into a coherent 3D representation, extracting anatomically meaningful features and subtle muscle tone changes.
- EEG-Level Accuracy: In validation experiments, Cheese3D predicted anesthetic depth by measuring facial muscle dynamics, matching the accuracy of invasive EEG methods while remaining non-invasive.
- Developmental and Clinical Potential: Because facial movement emerges early in development, this platform can help study how social communication matures and how conditions such as autism may disrupt that process.
Source: CSHL
Love, pain, joy, fear, desire — facial expressions carry the full spectrum of emotion. We interpret them almost instinctively, yet turning those impressions into precise, quantifiable data has proved difficult. Cheese3D narrows that gap by turning tiny, rapid facial movements into interpretable measurements linked to brain activity.
Cold Spring Harbor Laboratory (CSHL) Assistant Professor Xun Helen Hou and her team developed Cheese3D and describe it in a study published in Nature Neuroscience. The platform captures high-speed, face-wide 3D motion — including ears, eyes, whisker pad, and jaw on both sides — and converts those measurements into sub-millimeter, anatomically grounded metrics.
The need for this solution grew from the practical limits of current methods. Veterinarians and experienced researchers can often infer an animal’s state from its face, but there was no automated, reliable system that quantified those observations with enough resolution to reveal underlying neural processes. Mice are essential models in neuroscience, but their small, cone-shaped faces present unique imaging challenges. Cheese3D addresses those constraints by combining a multi-camera array with machine learning that fuses the views into a three-dimensional, time-resolved portrait of facial motion.
In experimental validations, the Hou lab used Cheese3D to monitor natural behaviors such as eating and to probe states like anesthesia. Working with the Borniger lab, they demonstrated that facial motion alone can indicate how deeply anesthetized a mouse is at a given moment. The non-invasive measurements matched gold-standard EEG readings while leaving the animal undisturbed, a major advantage for studies where natural behavior must be preserved.
“Very subtle changes in facial muscle tone teach us a lot,” Hou notes. By linking these subtle face-wide motions to neural recordings, Cheese3D provides a window into otherwise hidden brain and physiological processes.
Beyond anesthesia monitoring, the platform opens paths to clinical and developmental research. Facial movement appears early in life and is central to social communication; understanding how facial patterns develop could illuminate mechanisms relevant to autism and other behavioral disorders. The tool also helps reveal the mechanics of feeding, ear and jaw motion, and responses to targeted brain stimulation — including features measurable only in 3D, such as ear angles.
Key Questions Answered:
A: Mice do not express human-like smiles, but they display a rich repertoire of facial behaviors linked to pain, pleasure, stress, and general well-being. Veterinarians have long interpreted these signs; Cheese3D converts them into a quantitative, interpretable “dictionary” that maps facial movements to brain and physiological states.
A: EEGs require electrodes attached to the scalp or implanted, which can be invasive and alter behavior. Cheese3D provides a non-contact alternative that monitors brain-state correlates from a distance, preserving natural behavior and reducing procedural stress.
A: That is a long-term goal. Mapping how facial muscles correspond to neural circuits in mice can guide the development of non-invasive human tools for diagnostics, behavioral assessment, and safer monitoring under anesthesia.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The original journal paper was reviewed in full.
- Additional context and explanation were added by editorial staff.
About this neurodevelopment research news
Author: Samuel Diamond
Source: CSHL
Contact: Samuel Diamond – CSHL
Image credit: Neuroscience News
Original Research: Closed access.
Title: Cheese3D enables sensitive detection and analysis of whole-face movement in mice — Kyle Daruwalla, Irene Nozal Martin, Linghua Zhang, Diana Naglič, Andrew Frankel, Catherine Rasgaitis, Rubin Zhao, Xinyan Zhang, Zainab Ahmad, Jeremy C. Borniger & Xun Helen Hou. Nature Neuroscience
DOI: 10.1038/s41593-026-02262-8
Abstract
Cheese3D enables sensitive detection and analysis of whole-face movement in mice
Facial expressions and movements — from fleeting grimaces to rapid chewing — reveal moment-to-moment changes in neural and physiological state. Mice, which have distinct facial responses and conserved mammalian facial control circuits, serve as an ideal model to investigate how facial motion reflects underlying brain activity.
Existing methods have lacked the spatial and temporal resolution needed to capture the full range of mouse facial motion because of the face’s small, conical shape. Cheese3D overcomes these limitations with a calibrated six-camera array that records high-speed, face-wide 3D motion, covering ears, eyes, whisker pads, and jaw on both sides of the face.
The system extracts dynamics of anatomically meaningful 3D features in absolute world units with sub-millimeter precision. Proof-of-principle experiments demonstrate that Cheese3D can predict anesthetic depth from changing facial patterns, infer tooth and muscle anatomy from rapid ingestion movements, detect slight movement differences evoked by brainstem stimulation, and relate neural activity to spontaneous facial behaviors — including expressive features measurable only in 3D, such as ear-angle motion.
Cheese3D is positioned as a discovery tool that renders subtle mouse facial movements into a highly interpretable readout of otherwise hidden neural and physiological processes, advancing research into behavior, development, and clinical monitoring.