STIMscope (Spatio-Temporal Illumination Microscope) is an open-source benchtop platform for simultaneous imaging and patterned optical stimulation. A synchronized control system coordinates the camera, the DMD-based patterned-light projector, illumination, and GPU-accelerated analysis to support all-optical neural-interrogation experiments.
This repository packages the STIMscope platform as a Docker distribution for NVIDIA Jetson: the Qt GUI, the C++ projector engine, the calibration suite, the live-trace pipeline, hardware diagnostics, and the per-feature workflows so a complete setup can be reproduced on commodity edge hardware.
Reference: Chorsi et al., STIMscope: A high-resolution, low-cost, optogenetic stimulation platform for closed-loop manipulation of neural activity at the centimeter scale, bioRxiv 2026 — DOI 10.64898/2026.05.27.728160.
Fig 1a — Photo of the implemented STIMscope
platform in the inverted configuration: sample holder, objective,
GPU processing unit (NVIDIA Jetson AGX Orin), microcontroller, DMD,
and stage controller.
Each of the following is a first-class capability of the GUI — none is a prerequisite for the others, and the order in which an operator uses them depends on the experiment.
| Capability | What it does |
|---|---|
| Live camera acquisition | IDS Peak USB3 (default), MIPI, or generic camera; software or hardware trigger; analog + digital gain; per-frame exposure control |
| Recording | TIFF stacks of the live feed; snapshot for single frames; in-app + external TIFF viewers |
| DMD patterned projection | Send static masks, mask folders, or trial-driven sequences through the C++ projector engine; per-pattern trigger out for synchronization |
| Illumination control | DMD-internal RED/BLUE/RGB channels (DLPC3479 Illumination Select); GPIO camera + projector trigger lines via libgpiod (env-configurable) |
| Calibration suite | ArUco/ChArUco autonomous DMD→camera homography; Affine-SIFT feature-matching; structured-light sub-pixel LUT; reload/push existing H or LUT to the engine |
| Real-time trace extraction (RTTE) | Per-ROI mean fluorescence per camera frame; paginated multi-ROI plots in PyQtGraph; live ΔF/F overlay + optional OASIS preview deconvolution; snapshot + comprehensive export. |
| Offline ROI segmentation | Otsu (with optional watershed splitting); Cellpose (cyto2 / cyto / nuclei / custom) when installed. |
| Hardware diagnostics | Pixel probe, DMD R/B isolation, GPIO trigger pulse tests, engine monitor, LUT diagnostics suite (round-trip error, dot array, edge strip, calib characterization) |
| I²C control | Arbitrary DLPC3479 opcode bursts with templates, configurable bus + address, single-byte reads + atomic-burst writes |
| Sensor settings | Hardware-exposed analog gain, digital gain, exposure, contrast, gamma controls |
See Features and GUI Reference in the wiki for the full feature catalog.
The platform composes off-the-shelf parts into a bill of materials under USD $5,000 (preprint Abstract, Discussion). Synchronization between the image sensor, DMD projector, microcontroller, and Jetson follows the architecture in Fig 1b of the preprint.
| Component | What we use | Preprint reference |
|---|---|---|
| Compute | NVIDIA Jetson AGX Orin (JetPack 5/L4T R35.x or JetPack 6/L4T R36.x) | Methods § Image processing pipeline |
| Camera | Sony IMX334 / IMX290 small-pixel back-illuminated CMOS in an IDS Peak USB3 housing (MIPI / generic-camera paths also supported) | Methods § Camera; Fig 1b |
| Projector | TI DLP4710 DMD driven by DLPC3479 controller (I²C) | Methods § DMD; Fig 1b |
| Microcontroller | Microchip ATSAMD51 (Adafruit Grand Central M4) | Methods § Microcontroller; Fig 1b |
| Sync | libgpiod — gpiochip + line numbers env-configurable (STIM_GPIO_CHIP, STIM_CAM_LINE, STIM_PROJ_LINE) |
Methods § Synchronization |
The platform falls back to simulation-friendly modes (no camera, no projector) when hardware is absent — see Hardware Setup and Portability.
git clone https://github.com/Aharoni-Lab/STIMscope.git
cd STIMscope./build.sh # auto-detects JetPack version
export DISPLAY=:0
xhost +local:docker
sudo -E docker-compose up gui # full GUIFor prerequisites (NVIDIA Container Toolkit, IDS Peak SDK download path, JetPack-specific build args), see Install.
Every machine-specific value (data root, I²C bus, GPIO chip + lines, default fps/exposure, recording format) is an environment variable read at startup — no rebuild required to retarget a different Jetson or carrier board. See docs/PORTABILITY.md for the full env-var surface and a sanity-check on a fresh machine.
| Metric | Value | Preprint reference |
|---|---|---|
| Trigger-to-photodiode latency (mask → light) | 26.3 ms (mean) | Fig 4e; 5,000-mask photodiode run |
| End-to-end closed-loop latency (project + capture + ROI extract) | 91.6 ms | Fig 4f |
| Targeting accuracy (RMS error, ≈ 85,000 targets, 1936 × 1096 field) | 0.46 px ≈ 1.3 µm | Fig 4c |
| Imaging FWHM (lateral) | 5.6 µm center / 5.8 µm edge (4 µm fluorescent beads, f/4) | Fig 2c–e |
| Excitation FWHM (lateral) | 5.8 µm center / 6.2 µm edge (single DMD pixel) | Fig 2f–g |
| Field of view (demagnified) | 14 × 11 mm² | Fig 1f, Fig 3a |
The closed-loop end-to-end latency in Fig 4f explicitly excludes an inference model (preprint Discussion) — see docs/IMPLEMENTATION_NOTES.md for the scope and implementation status of the platform.
If you use STIMscope in research, see CITATION.cff (GitHub renders a "Cite this repository" button from it). The NOTICE file preserves upstream attribution. Figures reproduced in this repository are subject to the preprint's CC BY-NC-ND 4.0 license, independently of this repository's software license.
GPL-3.0 — see LICENSE.