UC Berkeley

The successful outcome of cancer therapy and treatment depends on the ability to ensure no microscopic residual disease is left behind during the treatment. It is therefore critical to be able to visualize and identify any residual disease being left behind in the tumor bed during resection surgeries, preferably in real-time and in an intraoperative setting. For practical purposes, leaving behind any more than 200 cancerous cells in the tumor bed increases the chance of cancer recurrence -across all cancer types-, highlighting the importance of microscopic residual disease.Recent advancements in fluorescently-tagged targeted molecular probes and imaging agents have enabled a significantly enhanced selectivity in detecting cancer cells, allowing single cell detection using conventional fluorescent imaging techniques. However, these techniques have remained largely impractical in intraoperative settings due to the fact that they rely heavily on large and cumbersome instruments, including rigid optical filters (for color and wavelength selectivity) and focusing lenses to be able to resolve the image from their operating distance. The bulky and rigid optical lenses and filters, required to resolve the weak fluorescence signal from background, are challenging to miniaturize, and restrict the imager to a relatively far working distance from the tumor cells, reducing both the sensitivity and maneuverability within complex tumor cavities. These limitations also impose a minimum size and form factor restriction on these optical imagers, precluding a majority of them from being deemed practical in surgical settings, and particularly hard to maneuver in today’s minimally invasive procedures. Ideal intraoperative molecular imaging platforms will not only require an ultra-small form factor sensor with high sensitivity (<1000 cells), but also the ability to perform deep tissue imaging to enable extracting valuable information from not just the tumor bed, but also the lymph nodes that are often sitting a few millimeters below the resection sites. This imager should be able to resolve images without any optics (focusing lenses, wavelength filters) and be compatible with a molecular probe that has the ability to be excited with a deep-penetrating wavelength such as in the near infrared (NIR) range. In this thesis, we shift the cumbersome optical requirements of fluorescence imaging into the time domain, using an optics-free micro-fabricated, time-resolved contact imaging array. Made possible through the synergistic integration of a custom high-speed integrated circuit imaging array and ultra-efficient upconverting nanoparticles (UCNPs), which are 2-3 orders of magnitude brighter than conventional lanthanide-doped UCNPs and have long (>100μs) phosphorescence lifetimes, this work provides an integrated and standalone imaging platform using a 200μm-thin imager, which functions as a scalable molecular imaging skin, able to be integrated on any surgical instrument, for real-time visualization of tumor cells in surgical settings. This sensor has been designed to be synergistically compatible with UCNPs, and their excitation wavelengths of either 980 nm or 1550 nm, for a 2-photon or 3-photon absorption mechanism respectively. This image sensor, measuring 4.8 mm by 2.3 mm, consists of an 80-by-36 pixel-array and includes custom-fabricated integrated micro-collimators to enhance resolution and sharpness when performing direct contact imaging, and by proxy obviating focusing lenses. To eliminate wavelength filters, a time-domain image resolution technique is implemented in which pulses of excitation light (980 nm or 1550 nm) with acquisition windows, taking advantage of the UCNP’s long emission lifetime. To mitigate the challenge of the effect of the background generated by the pulsed excitation light, we have implemented a novel dual-photodiode pixel architecture where a secondary diode is used to measure the local background level and later used to remove it from the main signal. We have optimized area consumption by the two in-pixel photodiode and maintained a fill factor of 47% and implemented a pixel-level cancellation scheme using a one-time and initial non-linear curve fitting to fine-tune the ratios of the two photodiodes. Using power levels compatible with in vivo use, we achieve cellular level detection and have validated the performance of this sensor using a 980 nm excitation with an intratumorally injected prostate tumor specimen in mouse and have been able to achieve a signal to background (SBR) level of 8 with an excitation power of 45 W/cm2 using pulsed 980 nm light. The imager is also able to reduce the background generated by the excitation light to single digit mV level, nearing its noise floor level of 2.2 mV rms. The imager developed in this thesis demonstrates the capability of optics-free CMOS imagers in being integrated in surgical settings and becoming miniaturized alternatives for large and cumbersome conventional fluorescent imagers.