Synthetic mesoscope to image large-scale neural activity with near-single-cell resolution in freely-behaving animals - Project Summary Most, if not all, cognitive functions require the coordination of multiple brain regions. The ability to simultaneously monitor neural activity across multiple brain regions in cellular resolution and high speed during natural behavior is important to study how different parts of the brain work together to support behavior. Doing so requires the recording device to have a high spatiotemporal resolution over a large field of view, and be sufficiently light and compact so it can remain stable on the animal’s head while the animal freely behaves. Optical methods are especially promising to fulfill all these requirements. However, existing optical miniaturized microscopes have tradeoffs among the field of view, resolution, and device footprint. To achieve fine resolutions, most of the existing miniaturized microscopes can only image a sub-millimeter scale field of view. We propose to develop a miniaturized imager that can record neuronal activity in near-cellular resolution over the entire cranium in freely-behaving mice. We have previously demonstrated a miniaturized 3D microscope where all bulk lenses are replaced by a thin layer of microlens array. The microlens array effectively splits the entire field of view into different regions, and uses different lens units to image them in parallel. A computational algorithm is then used to reconstruct and synthesize the 3D volume. This approach enables high-resolution imaging over a large 3D volume through a thin device. Our microscope achieved <10 µm lateral resolution over 4x6 mm2 field of view, and can reconstruct 3D volume across ~600 µm depth range with ~50 µm axial resolution from a single 2D captured image. In this project, we will optimize and expand the microlens array so it could image neural activity across the entire cranium (~9x10 mm2) and 400 µm depth in near-cellular resolution in freely-behaving mice. We will package the image sensor, microlens array, and illumination optics into a compact device. We will develop deep-learning-based algorithms that can faithfully reconstruct neuronal activity over the 3D cortical volume from recordings that are highly corrupted by light scattering in real time. Compared to existing algorithms, our approach can significantly reduce the required computational resources and processing time. Our miniaturized imager enables the investigation of very large-scale neuronal circuits distributed across the cortex. We will run testbed experiments on both freely-moving mice and head-unrestrained macaque monkeys, two important animal models for studying the neural basis of behavior. We will leverage well-established experimental paradigms, including passive viewing of visual stimuli and navigation in open arenas. These experiments will serve to validate, benchmark, and further refine our techniques in vivo. The cross-species experiments will not only prove the potential of our imager, but also provide critical insights into how animal behavior emerges from the orchestration of individual neurons in large-scale neural circuits. This development and validation of this new technology will pave the way for many exciting future experiments, and greatly enhance our ability to address important neuroscience questions that are challenging with the existing techniques.
