Portable Integrated Fourier Ptychographic Microscope

NASA’s integrated, portable FPM device combines advanced optical microscopy with AI in a compact form factor. At its core, the system uses a Raspberry Pi camera module equipped with an 8-megapixel sensor and a 3-mm focal-length lens, achieving approximately 1.5× magnification. Illumination is provided by a Unicorn HAT HD LED array positioned 65mm below the sample stage, creating a synthetic numerical aperture of 0.55. All these components are controlled by an NVIDIA Jetson Nano board, which serves as the system's embedded AI computing platform. NASA’s portable FPM device can be integrated with a microfluidic system for sub-micron imaging of liquid samples. In addition to its portability, what sets NASA’s integrated FPM device apart is its integration of deep learning capabilities. The invention has two modes – normal mode and deep learning mode. In its normal mode, the system captures data and performs intensity and 3D phase analysis using traditional FPM methods. The deep learning mode enhances this base functionality by employing neural networks for image reconstruction and system optimization. The AI system automatically detects when samples are out of focus and can either mechanically or digitally adjust to the correct focal plane. To achieve near real-time monitoring, deep learning models significantly reduce data acquisition time by selectively using only a portion of the LED array and provide fast reconstruction capabilities leveraging training on specific sample types. While originally designed for imaging biosignature motility in liquid samples for spaceflight applications, this NASA innovation can image many different types of samples, and is not limited to biological specimens. The ability to operate this system in the field further broadens use-cases. Fourier ptychographic microscopy (FPM) seeks to address the tradeoff between resolution and field-of-view (FoV) experienced by conventional imaging methods. The technique combines multiple low-resolution images captured under varying illumination angles to reconstruct a single, high-resolution, wide FoV image. Key challenges in FPM devices include lack of portability (preventing use in the field) and long image reconstruction times (preventing real-time feedback, decreasing throughput and efficiency, and leading to poor user experience that presents a barrier to adoption). Originally designed for biosignature detection and motility in liquid samples, NASA’s Goddard Space Flight Center has developed a compact, portable inferencing device comprised of a FPM integrated with a microfluidic device to perform wide FoV, sub-micron spatial resolution imaging of several different sample types. NASA has significantly improved FPM device technology by creating a small portable device that leverages algorithms to perform self-calibration of LED positions and trained, sample type-specific neural networks for recording and image reconstruction. These algorithmic optimizations allow for near real-time image reconstruction.