My research interests are in custom imaging sensors, especially compact, polarization sensistive sensors. I have worked to not only develop and test the hardware for these sensors, but also to apply them to solve problems in the fields of neuroscience, tissue analysis, and biomedicine. Below is a summary of some of the work I have done:
A major challenge in neuroscience today is measuring large populations of neurons simultaneously across a wide spatial area in-vivo. The most common method uses fluorescent markers or other voltage sensitive dyes which excite during neural activation. While effective, these methods can be potentially toxic, while also photobleaching over time, limiting the potential for long term recordings. A potential alternate optical method detects the intrinsic changes of the neural cells' activity by detecting the polarization properties of reflected light without using any invasive dyes or makers. To detect these changes requires a compact polarization sensor which operates in real-time. I have fabricated such a custom sensor using a 180nm process, desining both the underlying current mode CMOS imager, as well as integrating the pixel-matched polarization filters. The custom sensor was used to image intrinsic neural activity in-vivo from the antennal lobe of a locust exposed to the odors hexanol and octanol. The sensor measured a change in polarization response of 0.38% over the baseline for hexanol, 0.15% over the baseline for octanol, and 0.45% over the baseline for the combined odors, which demonstrated for the first time the potential use of polarization change as a method of neural capture.
Polariztion imaging is commonly used to study the direction and strength of fiber alignment in soft tissues. These methods typically involve placing the tissue sample under a known load, and spinning linear polarization filters while the tissue is static. While effective, these methods have so far been unable to measure the real-time dynamic loading of tissue. To solve this, I developed a new optical system that can measure tissue dyanmics in real-time and at high resolutions. The technique uses a Division-of-Focal-Plane image sensor in conjunction with circularly polarized light trasnmitted through the tissue sample. The technique produces two-dimensional alignment and strain maps under any dynamic loading conditions in real-time and without moving optics, allowing for more complex loading protocols and analysis techniques.
Compact, high-resolution Division-of-Focal-Plane sensors are a recent development. The evaluation of their performance may not be well understood, as both the integrated optics and underlying image sensor can influence the measurement of the polarization state of a scene. Factors like the wavelength of light detected, the contrast ratio of the pixel-matched filters, and even the lens used can all have an impact. To quantify exactly how these different factors alter the measurement, I have developed a systematic testing methodology which incorporates multiple states of polarization, wavelengths, and light angles for a thorough evaluation of a sensor's performance.
The typical way to measure the polarization of a scene is by rotating a filter at different orientations and capturing multiple images of the scene. This method, while simple, requires a static scene and cannot be used for real-time measurements. Advances in fabrication technolgy have allowed polarizers the size of pixels to be deposited directly on top of an image sensor, enabling the instantaneous capture of the polarization of a scene without moving optics. I have helped to develop a high frame rate polarimeter using this method. The sensor can image at 250 frames-per-second at VGA resolution, and over 1000 fps when limited to a subsection of the whole array.
Division-of-Focal-Plane polarimeters allow real-time capture of the polarization state of light in a scene. As these sensors can contain large spatial resolutions and high frame rates the image processing to compute and display the polarization becomes a potential bottleneck when using these sensors in the field. To solve this, we implemented the polariztion image processing steps of per-pixel calibration, spatial interpolation, and degree and angle of polarization computation on a multi-core general purpose CPU architecture, GPU, and FPGA. We evaluted each platform's performance in terms of throughput (frames-per-second), power consumption (W/frame), precision (floating-point-error), and physical limitations.