Near-Infrared Optical Tomography with Single-Photon Avalanche Diode Image Sensors

In the past decades different methods based on magnetic resonance imaging (MRI) and positron emission tomography (PET) have been developed to image a variety of hemodynamic parameters in tissue. However, simultaneous high resolution imaging of oxy- and deoxyhemoglobin is not yet possible in clinical practices with millimeter accuracy. Near-infrared Spectroscopy (NIRS) is a well established technique that has been widely utilized in research and clinical practice to monitor the concentration of different chromophores in blood. It is capable of measuring a variety of parameters such as oxygenation levels in tissue, hematocrit, and cytochrome c oxidase levels among others. By increasing the amount and complexity of the light sources and detectors integrated in NIRS systems, in the last years researchers have been able to acquire tomographic images of such chromophores. This new imaging modality is known by different names: near-infrared optical tomography (NIROT), optical tomography (OT), diffuse optical tomography (DOT), or diffuse optical imaging (DOI). One of the main factors that limited the quality of the tomographic images obtained with NIROT has been the low number of sources and detectors employed in NIROT systems. The integration of the first single-photon avalanche diode (SPAD) together with time-to-digital converters (TDCs) in CMOS enabled a whole new range of possibilities in the field of single-photon detection. In this thesis the application of a SPAD-TDC image sensor in NIROT is presented for the first time. The main objective was to develop a new system that could perform acquisitions nearly in real time and that was capable of delivering tomographic images in a short period of time for medical evaluation. A new optical setup was conceived based on this detector to take advantage of the large amount of information delivered by the SPADs. By employing line-shaped illumination sources instead of point like sources, the target is more homogeneously illuminated and consequently a reduced number of sources is necessary. It was experimentally demonstrated that a resolution of 5 mm is possible with this new NIROT system. New algorithms that reduced the ill-posed nature of the image reconstruction problem were developed thanks to the wide-field time-resolved measurements delivered by the SPAD image sensors. The large datasets obtained with our NIROT system and its time-resolved capabilities enabled the development of fast methods capable of reconstructing an image with millimeter resolution in a few seconds with a laptop computer. We also analyze the effect of microlenses on the light sensitivity of the image sensor, showing that it is possible to increase it by a factor of 10 under certain conditions. A new image sensor with 4x400 pixels implemented in a 3D CMOS technology for NIROT applications is also presented. To our knowledge, this is the first backside illuminated SPAD array that incorporates in-pixel TDCs. A novel TDC architecture was introduced that reduces the energy consumed per conversion, consequently allowing a high number of TDCs working in parallel. The study of how NIROT can benefit from SPAD image sensors is presented in this thesis. Despite their current limitations, they enable the implementation of systems with thousands of detectors capable of millimeter resolution. Further developments in SPAD array architectures and TDCs will continue improving the performance of time-resolved NIROT systems.

Single-Photon Techniques for Standard CMOS Digital ICs

Claudio Favi

The advent of single-photon detectors known as Single-Photon Avalanche Diodes in standard CMOS technology opened the way to new perspectives in integrating these ultra sensitive light sensors with digital logic. Light has some interesting properties that attracted researchers in computer and electronics for a long time. Its weightlessness nature makes it a candidate to replace electrons when it didn't already do so. This is particularly true for long distance data transfers. However a new trend can be observed. Researchers are looking into photonics, the science of light, for short distance communications as well. Power efficiency is one of the reasons of this trend. In fact, photons, the elementary particles of light, don't suffer of resistive, capacitive, or inductive effects like electrons do. The wavelike nature of light can also free designers from problems such as cross-talk and side-channel noise injection while taking advantage of interference. However photonics is still not the panacea and we don't believe it will ever completely replace electronics. Most certainly a combination of the two will be adopted to take advantage of both worlds. CMOS digital Integrated Circuit (IC) design faces many challenges and it is not clear whether technology will still provide small footprints, high performance, and low power in the future. Photonics with SPADs can answer some of these questions. In this work, we present three investigations where SPAD photonics is integrated within digital CMOS technology. The main contributions of this thesis are threefold: single-photon CMOS communication paradigms, single-photon processing and readout techniques, single-photon clocking and synchronization methods. Single-photon communication was achieved using a combination of SPADs and ultra-fast TDCs in a pulse position modulation scheme. In this context, theoretical channel capacity limits in the presence of noise and other non-idealities typical of SPADs were derived; a TDC with a resolution of 17 ps was demonstrated in a standard FPGA fabric. To the best of our knowledge, at the time of this writing, this is the highest reported resolution for a TDC of this kind. Single-photon processing and readout was achieved in several technologies, focusing on image sensor design, whereby massive parallel architectures were studied and implemented in CMOS. Single-photon clocking and synchronization was demonstrated allowing potentially zero skew systems irrespective of the chip area. The power benefits of this approach in embedded systems with instruction set extensions are particularly interesting. The thesis makes use of SPAD technology implemented in CMOS for a number of applications creating a bridge between digital design and high performance photonics. We believe that this is the first attempt in this direction focused on CMOS technology.

EPFL2011

Arrays of Single Photon Avalanche Diodes in CMOS Technology: Picosecond Timing Resolution for Range Imaging (INVITED)

Pierre-André Besse, Edoardo Charbon, Cristiano Niclass

A solid-state imager fabricated in CMOS technology is presented for depth information capture of arbitrary 3D objects with millimeter resolution. The system is based on an array of 32x32 pixels that independently measure the time-of-flight of a ray of light as it is reflected back from the objects in a scene. A single cone of pulsed laser light illuminates the scene, thus no complex mechanical scanning is required. Millimetric depth accuracies can be reached thanks to the rangefinder’s optical detectors that enable picosecond time discrimination. The detectors, based on a single photon avalanche diode operating in Geiger mode, utilize avalanche multiplication to enhance light detection. Optical power requirements on the light source can therefore be significantly relaxed. A number of standard performance measurements, conducted on the imager, are discussed in this paper. The 3D imaging system was also tested on real 3D subjects demonstrating the suitability of the approach.

2005

Mathematical Modeling of Brain Energy Metabolism, Measured with PET and MRS in Rodents

Bernard Lanz

Positron emission tomography (PET) and nuclear magnetic resonance spectroscopy (MRS) are two biomedical measurement techniques developed in the end of the XXth century, which drastically improved the amount of accessible information available in vivo. PET became popular through the most widely used tracer, fluorodeoxyglucose (FDG), which enables the measurement of the local glucose utilization and is nowadays routinely applied in clinical practice. Nuclear magnetic resonance is used in a large array of applications such as in analytical chemistry or chemical structure determination and is essentially known for its versatile medical imaging capabilities, grouped under the name of magnetic resonance imaging (MRI) and magnetic resonance spectroscopy (MRS). In vivo MRS measures concentrations of metabolites by interacting with various nuclei such as 1H, 13C or 15N, with the major asset that MRS enables the identification of the chemical position at which the detected nuclei is located in the measured molecule. When coupled with labeled substrate infusions, dynamic MRS gives the opportunity to probe specific biochemical reactions directly in vivo, which opens the way to a wide range of metabolic studies. However, the correct interpretation of the elaborate dynamic labeling data acquired either with MRS or PET tracer experiments requires the application of adapted metabolic models to derive quantitative metabolic rates characterizing the biochemical processes under study. The work of this thesis involves both PET and MRS studies of brain energy metabolism in rodents and focuses on the development of adapted metabolic models to derive reliable metabolic fluxes characterizing various brain metabolic processes, such as glucose consumption, glial and neuronal oxidative metabolism and neurotransmission. FDG-PET enables the measurement of the cerebral metabolic rate of glucose (CMRGlc) using a three-compartment metabolic model for FDG transport across the blood-brain barrier and FDG phosphorylation to FDG-6P. Although this method is well established, a main drawback of CMRGlc measurement with FDG-PET is the necessity to characterize the available FDG concentration in the plasma over the experiment duration, the so-called arterial input function. This constraint is a strong limitation for preclinical studies in rodents, due to their low blood volume. We developed therefore a new method to extract the FDG input function directly from the PET images in rats and mice, using the time-activity curve of a voxel located in the inferior vena cava. The method was validated by comparison with a dedicated external blood counter and manual blood sampling. Using this method, a CMRGlc of ~0.22 μmol/g/min was determined in the mouse cortex. Glial oxidative metabolism can be specifically assessed using glial specific substrates, in particular [1-11C] acetate. However, no existing metabolic model described the 11C tissue activity curve in terms of neuroglial energy metabolism. In order to extract quantitative metabolic fluxes characterizing the glial TCA cycle rate and glutamate/glutamine cycling, we adapted the neuroglial two-compartment modeling approach previously used in 13C MRS cerebral metabolic studies to interpret positron emission data following 11C-acetate injection in the rat. The precision and accuracy of the estimated metabolic parameters was tested and a composite glial metabolic flux Vgtg=0.136 ± 0.042 μmol/g/min and a neurotransmission flux Vnt=0.170 ± 0.103 μmol/g/min were obtained. This approach enabled a direct comparison of the metabolic fluxes measured using 11C PET with the values determined with 13C MRS. In the field of 13C MRS, we took advantage of the higher sensitivity and spectral resolution available at high magnetic field (14.1T) to measure glial and neuronal oxidative metabolism, glutamatergic neurotransmission and pyruvate carboxylation in more details and with higher precision using both [1,6-13C2] glucose and [2-13C] acetate infusions in rats. In the acetate study, the glial specific uptake of acetate enabled a quantification of glial oxidative metabolism with unrivalled precision. We could estimate for the first time separately the transmitochondrial flux Vx in the glial and neuronal compartment, a biochemical pathway which has been subject of strong controversy. The glial and neuronal Vx were determined with high precision, as calculated through Monte Carlo simulation, and their value was on the same order of magnitude than the respective glial and neuronal TCA cycle flux. The model was extended to estimate the fraction of glutamate located in the glial compartment. A glial glutamate concentration of 0.6 ± 0.1 μmol/g was found. The values of the different metabolic fluxes characterizing the neuroglial system (Vtcag=0.27, Vxg=0.17, VPC=0.087, VNT=0.15, Vtcan=0.37 and Vxn=0.46 μmol/g/min) were in very good agreement with the values found using [1,6-13C2] glucose infusion. In addition, the effect of the MRS temporal resolution, experiment duration and signal- to-noise ratio on the precision of the derived metabolic fluxes was analyzed. Previous simulation studies showed that two-compartment modeling of neuroglial metabolism using 13C glucose infusion could lead to unreliable parameter estimations. However, using polarization transfer 13C MRS at 14.1T, the carbon positions C4, C3 and C2 of glutamate and glutamine as well as the positions C3 and C2 of aspartate could be measured with high SNR. The C2 positions were used in two-compartment modeling for the first time, enabling a precise measurement of the apparent glutamatergic neurotransmission and glial metabolism, including pyruvate carboxylation, as confirmed by Monte Carlo simulation. The glial dilution at the level of acetyl-CoA, which is related to the metabolism of other substrates than glucose, could also be characterized. Brain metabolism under hyperammonemic conditions was probed with 15N MRS under labeled ammonia infusion and modeled using a modified neuroglial two- compartment model describing the dynamics of [5-15N] glutamine, [2-15N] glutamate and [2-15N] glutamine. Due to detoxification processes, the total glutamine concentration, measured with 1H MRS, was steadily increasing. This non-steady- state metabolic condition resulting from the hyperammonemic stress was taken into account in the model, resulting in the determination of the apparent neurotransmission (0.26 ± 0.030 μmol/g/min), the glutamate dehydrogenase (0.029 ± 0.002 μmol/g/min) and net glutamine accumulation (0.033 ± 0.001 μmol/g/min). Finally, compartmental modeling was applied to 13C labeling studies in the awake rat, fed during 5, 24 or 48 h with a [1-13C]-labeled glucose solution, to determine brain glycogen and NAA turnover times (τGlyc=5.3 ± 3.2 h and τNAA=15.6 ± 6.5 h), using 13C spectra of brain extracts. A group of rats followed a mild brain activation protocol over 5 hours, which resulted in a decreased glycogen turnover time (τGlyc =2.9 ± 1.2 h).