This conference presentation was prepared for SPIE BiOS 2023

We report one of the first studies on direct 3-D printing of heterogeneous optical phantoms with programmable absorption and scattering properties using a multi-color mixing extruder. This method dynamically mixes off-the-shelf gray, white, and translucent filaments to achieve arbitrary target absorption and scattering coefficients. We use a spatial frequency domain imaging system to characterize and validate the printed properties and verify that they follow our hypothesized linear-mixing models. A complex phantom with five inclusions with distinct optical properties was produced and the measured properties compared to their predicted values showed an error between 12%-15%.

Structured interrogation (SI) is a frequency-domain NIRS technique that adjusts the relative phase between intensity-modulated light sources and generates a various spatial pattern of depth sensitivity. It can be used to extract quantitative information from multi-layered tissues. We apply Cramér-Rao lower bound (CRLB) analysis for phase selection. Then, the selected SI phases are used to resolve the optical properties of a fat/muscle tissue model. We found that CRLB is effective for selecting SI phase shift, and a combination of SI measurements and multi-distance measurements can result in an accurate estimation of optical properties and allow for more compact fd-NIRS probes.

In this work, a miniaturized heterodyning FD-NIRS instrument was presented. The device uses a dual slope probe, which removes the need for pre-calibration. The low-footprint system consists of only the circuit board, the probe and a Raspberry-Pi. Four lasers (685 nm and 830 nm) and two avalanche photodiodes are used, where the lasers are modulated with 80 MHz and the APD signals are amplified and downconverted by the analog front end of the instrument: a custom designed fully differential ASIC in 130 nm CMOS technology. Solid phantom measurements revealed <9% error and significant stability for long-term measurements.

We propose novel frequency-domain data types, that show hybrid features to those of phase and AC intensity, i.e., better CNR features than phase data while preserving preferential sensitivity to deep tissue regions (like phase data). We show the CNR features of some of the novel data types in the dual-slope source-detector arrangement in the semi-infinite homogeneous medium and in the two-layer geometries. The results show that these novel data types indeed may have some hybrid features of AC and phase and may have a potential application in imaging of tissue.

Current Time-Resolved Near-Infrared Spectroscopy (trNIRS) fitting methods require the inclusion of an amplitude term, to account for the unknown gain in the measurements. This increases the number of fitting parameters and the potential for crosstalk between them. We propose a method that eliminates the need for the amplitude term by fitting the temporal derivative of the natural log of the distributions of time-of-flight. The new approach was tested on a large in silico dataset and the accuracy of the estimated cerebral oxygenation and total hemoglobin were 2.2±2.4% and 1.9±0.8%, respectively. Future work will include further validation with in vivo data.

Many studies on diagnosing adult chronic diseases such as diabetes have been achieved by analyzing blood data. Here, we present machine learning algorithm-based diagnostic methods for diabetes by analyzing blood flow oscillations. We used diffuse speckle contrast analysis(DSCA) to measure the blood flow of rats. It can non-invasively measure changes in the relative blood flow of the tissue. Blood flow data acquired from Streptozotocin-induced and control rats were preprocessed by wavelet transform and then classified from machine learning algorithms. In conclusion, the machine learning method can successfully classify two blood flow oscillations in diabetic and control rats.

Ultrasound (US)-guided diffuse optical tomography (DOT) has demonstrated potential for breast cancer diagnosis. Previous diagnostic strategies all require image reconstruction, which hindered real-time diagnosis. In this study, we propose a deep learning approach to combine DOT frequency-domain measurement data and co-registered US images to classify breast lesions. The combined deep learning model achieved an AUC of 0.886 in distinguishing between benign and malignant breast lesions in patient data without reconstructing images.

An intuitive and generalisable approach to spatial-temporal feature extraction for brain-computer interface (BCI) with high-density functional near-infrared spectroscopy (fNIRS) data is proposed, demonstrated here with frequency-domain (FD) signals for motor-task classification. Statistical analysis of the results shows that the spatially resolved convolutional neural network (CNN) model improves classification accuracy by 2.5% compared to a standard temporal CNN, further enhanced by data availability. This is a significant improvement considering the requirements of real-time BCI.

NIRS measurement is known to be susceptible to motion artifact, instrumental noise, etc. When analyzing NIRS time series, we almost always lack the ground truth for evaluation, and from many methods, it could be hard to pick the most appropriate method for one’s application. In this work, we proposed and examined the following pipeline: First, generate ground truth synthetic NIRS signal. Second, add application specific noises/activations. Finally, train a Long Short-Term Memory deep learning model for time series prediction. Our results revealed that this approach can recover the uncorrupted NIRS/PPG signal accurately and efficiently from noisy signals.

Diffuse optical spectroscopy is a widely used method for the non-invasive recovery of important biological factors such as tissue oxygenation and total hemoglobin concentration. Frequency domain-diffuse optical spectroscopy improves the accuracy of parameter recovery over continuous wave-diffuse optical spectroscopy by enabling the decoupling of tissue absorption from tissue scattering. However, this comes at the price of increased instrumentation cost and complexity. Here we detail an easy to build, low-cost, and robust frequency domain-diffuse optical spectroscopy system to increase accessibility to this technology along with testing of the system’s stability and accuracy to ensure its applicability for biological measurements.

We present a novel system based on a four-stage fiber delay network designed for multistate time-domain diffuse correlation spectroscopy, providing three output fibers per each delay state. The fiber delay network is coupled to a custom pulsed laser at 1064 nm and four SNSPDs, allowing to measure up to 12 independent source-detector pairs simultaneously. The system delivers 300ps optical pulses, 100 mW average optical power per fiber output, operates at 62.5 MHz and each cycle provides 4 laser pulses displaced of 4 ns. The instrument has been validated on healthy human subject during functional tasks, proving state-of-the-art performance.

Infants born at an extremely low gestational age are at an increased risk of intraventricular hemorrhaging during the first three postnatal days. We have built a standalone easy-to-use multi-wavelength multi-distance diffuse correlation spectroscopy system, which utilizes three time-multiplexed long coherence lasers at 785, 808, and 853 nm, single photon detectors, and photon time-tagging electronics to simultaneously quantify cerebral blood flow, tissue optical properties, and blood oxygen saturation. The system has been designed specifically for use on preterm infants. The device shows good agreement with a commercially available NIRS-DCS system. We are currently monitoring preterm infants and will show results.

Interferometric DOS (iDOS) is a new class of approaches that promises to improve the quantitative accuracy and depth specificity of blood flow index (BFI). iDOS techniques have alternatively achieved either time-of-flight (TOF) discrimination or highly parallel detection, but not both at once. Here, we break this barrier with a single iDOS instrument. Specifically, we show that rapid tuning of a temporally coherent laser during the sensor integration time increases the effective linewidth seen by a highly parallel interferometer. With a deep TOF filter applied to a high throughput interferometer, we demonstrate brain-specific BFI imaging.

Extracorporeal membrane oxygenation (ECMO) is an invasive technique that is used for patients that have experienced heart or lung failure. Due to the complex nature of this treatment, neurological injuries such as hypoxic ischemic encephalopathy (HIE) or stroke can occur. Conventional neuroimaging is risky for this patient population and only provides a snapshot in time, so continuous non-invasive bedside neuromonitoring. In this study we plan on monitoring cerebral blood flow using diffuse correlation spectroscopy and neural activity with electroencephalography. Correlating blood flow with neural activity will provide insight into the neurovascular coupling between neurologically injured versus normal patients on ECMO.

Interpretation of cardiac pulsatility in cerebral blood flow measured with transcranial doppler ultrasound (TCD) plays an important role in neurocritical care. However, TCD is only sensitive to large arteries, whereas the optical technique diffuse correlation spectroscopy (DCS) can detect microvascular flow. We develop and apply analysis methods to quantify morphological features of cerebral blood flow waveforms measured with DCS. We characterize normative waveforms in healthy adults at rest and report a waveform response to an intervention known to induce vasodilation (hypercapnia). We further show changes in waveform morphology in subarachnoid hemorrhage (SAH) patients with cerebral vasospasm compared to healthy controls.

Ultrasound (US)-guided diffuse optical tomography (DOT) has demonstrated success in breast cancer diagnosis. However, DOT data pre-processing and reconstruction still require some level of manual operation, for example, contralateral reference selection and data cleaning. In this study, we introduce an automated data pre-processing and reconstruction pipeline to accelerate the DOT clinical translation. The pipeline has integrated several data pre-processing modules and reconstruction methods that are adapted to data. The pipeline is implemented using a graphical user interface. Initial testing has shown that it can automate DOT right after the data acquisition and provides an accurate diagnostic score on cancer vs. benign probability.

Multimodal x-ray mammography and optical imaging data were acquired on six breast cancer patients who underwent neoadjuvant chemotherapy (NACT) but reponded differently to their treatment. Changes in tumor contrast quantified by total hemoglobin concentration (HbT) between baseline and pre-cycle 3 are distinctive across various levels of pathological outcomes. While decreases in lesion size have been observed in all cases regardless of pathological outcomes, optical contrast shows more distinctive response characteristics that could potentially be used to differentiate complete responders from partial responders.

The goal of this study was to analyze temperature dependent changes in NIR reflectance of ex vivo freshly excised human skin. A broadband halogen lamp was used as a light source. Remitted light was collected using an optical fiber and analyzed by a grating spectrometer. Samples were heated to and maintained at temperatures up to 60°C using a temperature control unit. The temperature was also independently monitored using a thermal camera. Consistent changes found include blue shift in water absorption peaks. This method may prove to be useful for monitoring tissue temperature during clinical procedures in real time.

We present on blood optical property alterations induced by lipids. Mie simulations were conducted to estimate the magnitude of μ_s^' changes due to changes in lipoprotein particles in blood after a meal. Longitudinal SFDI measurements were performed on the dorsal surface of volunteers’ hands pre and post high fat meal for 5 hours to monitor optical property changes within superficial vessels. The results show an increase in μ_s^' and a decrease in μ_a with higher changes observed in SFDI measurements compared to Mie simulations, potentially due to hemodynamic alterations that occur after a meal.

Optical technologies are widely used for tissue sensing purposes, however maneuvering conventional probe designs with flat-tipped fibers in narrow spaces can be challenging, such as in pelvic colorectal cancer surgery. In this study, a compact side-firing fiber probe was developed for tissue discrimination during colorectal cancer surgery using diffuse reflectance spectroscopy. The light behavior was compared to flat-tipped fibers using both Monte Carlo simulations and the tissue classification performance was examined using freshly excised colorectal cancer specimens. Using the developed probe and classification algorithm, we achieved an accuracy of 0.92 for the discrimination of colorectal tumor tissue from healthy tissue.

Jones fractures of the proximal fifth metatarsal carry a significant risk of nonunion, which greatly increases the burden on patients. A significant reason for this fracture’s poorer outcomes is the lack of blood flow in the region it occurs. We will apply our newly developed co-registered speckle contrast and frequency domain-diffuse optical tomography system to monitor the Jones fracture region in healthy subjects under and over 50 years of age. The information collected will provide baseline data for future fracture healing outcomes research and provide insight into whether the poorer healing outcomes of older individuals are related to baseline hemodynamics.

To accurately reconstruct fluorophore concentrations in-vivo in fluorescence lifetime-based tomography, it is necessary to accurately model and optimize, both the propagation of light in tissue in complex geometries and the system parameters involved in the entire optical chain from the source to the detector. This paper proposes a comprehensive methodology for accurately modelling a time domain tomography system along entire the optical chain, including the time-gated intensifier. Based on this modelling, we optimize the system parameters to obtain signals with high SNR at a faster acquisition rate. We validate the model using simulations and in-vivo experimental studies.

Programed death ligand-1 (PD-L1) expression is currently the only predictive biomarker for cancer immunotherapy. Since PD-L1 expression in tumors is largely heterogeneous, in-vivo detection and quantification of PD-L1 in intact tumors is of major interest. Here we employ fluorescence lifetime (FLT) imaging for in-vivo detection and quantification of PD-L1 expression using an anti-PD-L1 antibody conjugated to IRDye800CW (αPDL1-800). We show that FLT imaging accurately identifies heterogeneous PD-L1 expression in tumors. Tumor areas of high PD-L1 levels were spatially correlated to significantly longer FLTs of αPDL1-800 and the distribution of PD-L1 in deep-seated (>1cm depth) tumors was achieved using FLT tomography.

We previously introduced a novel method called temperature modulated fluorescence tomography (TMFT), which utilizes temperature sensitive ICG loaded pluronic nanocapsules termed ThermoDots and high intensity focused ultrasound (HIFU). TMFT leverages the superior sensitivity of the fluorescence imaging and the high spatial-resolution of focused ultrasound. Previously, we have presented a prototype system using a CCD camera. However, the scan time was too long for in vivo imaging since HIFU was scanned in a step-and-shoot mode. Here, we present the new continuous line scanning scheme, which drastically reduces the imaging time and paves the wave for in vivo preclinical imaging.

Bioluminescence tomography (BLT), as an available image-guided system for pre-clinical radiotherapy research, can localize in vivo tumors within high localization accuracy but it is still challenging to recover accurate structure information due to optical diffusion and ill-posed inverse problem. Recognition of this challenge led us develop novel reconstruction method, optimized spectral-derivative compressive sensing conjugate gradient algorithm. We will perform simulation and in vivo experiments to test BLT’s performance in reconstructing the target location and shape both in primary tumor or metastatic setting. We expect that our BLT-guided system will provide investigators quantitative tumor imaging to perform high precision radiation research.

The challenge to study radiotherapy for pre-clinical pancreatic ductal adenocarcinoma(PDAC) is lack of image-guided system providing strong soft-tissue contrast. Bioluminescence imaging is an attractive solution, but it is inadequate to localize movable PDAC at single projection. With these targeting uncertainties, large radiation beam is commonly used. Therefore, we innovated bioluminescence tomography(BLT) to guide conformal irradiation for PDAC. Our BLT could retrieve orthotopic PDAC volume within 0.5-2mm accuracy at 2-6mm depth, even tumor located at various locations at given study days. Our BLT offers unique opportunities to localize PDAC for conformal irradiation, reduce normal tissue toxicity, and therefore increase study reproducibility.

Neurofeedback addresses a control problem where the controller cannot directly manipulate nor observe the dynamic process to be controlled. This work proposes a new multiple-input single output fuzzy controller to induce designed sensorimotor modulation. A multivariate error function guides: the plant response (behavioral output) and the plant structure (brain organizations), targeting a functional reorganization strategy encoded as a graph. The model was tested on a synthetic human agent (plant) observed with fNIRS, considering the common physiological noise models and a force transducer (behavioral). The proposed fuzzy controller modulates brain connectivity to the targeted with repercussions for neurorehabilitation and neurofeedback applications.