Image Reconstruction Using Deep Learning for Near-Infrared Optical Tomography: Generalization Assessment.

Time is one of the most critical factors in preventing brain lesions due to hypoxic ischemia in preterm infants. Since early detection of low oxygenation is vital and the time window for therapy is narrow, near-infrared optical tomography (NIROT) must be able to process the high-dimensional data provided by today's advanced systems in the shortest possible time. Deep learning approaches are attractive because they can exploit such high information density while reducing inference time. The aim of this study was to evaluate the performance of a hybrid convolutional neural network, designed for NIROT image reconstruction and trained on synthetic data. Generalization capability was assessed using measurements on phantoms of a surface topology more divergent than the range of variation in the geometries of the in-silico data, with unseen, non-spherical inclusion shapes, and with source and detector arrangements different from those used for data generation. Substantial gains in speed, localization accuracy, and high image quality were achieved even under the highly varied measurement conditions.

Time Domain Near-Infrared Optical Tomography Utilizing Full Temporal Data: A Simulation Study.

The analysis of full temporal data in time-domain near-infrared optical tomography (TD NIROT) measurements enables valuable information to be obtained about tissue properties with good temporal and spatial resolution. However, the large amount of data obtained is not easy to handle in the image reconstruction. The goal of the project is to employ full-temporal data from a TD NIROT modality. We improved TD data-based 3D image reconstruction and compared the performance with other methods using frequency domain (FD) and temporal moments. The iterative reconstruction algorithm was evaluated in simulations with both noiseless and noisy in-silico data. In the noiseless cases, a superior image quality was achieved by the reconstruction using full temporal data, especially when dealing with inclusions at 20 mm and deeper in the tissue. When noise similar to measured data was present, the quality of the recovered image from full temporal data was no longer superior to the one obtained from the analysis of FD data and temporal moments. This indicates that denoising methods for TD data should be developed. In conclusion, TD data contain richer information and yield better image quality.

BIAN: A Multilayer Microfluidic-Based Tissue-Mimicking Phantom for Near-Infrared Imaging.

Near-infrared spectroscopy (NIRS) is a non-invasive optical method for monitoring cerebral oxygenation. Changes in regional blood flow and oxygenation due to neurovascular coupling are important biomarkers of neuronal activation. So far, there has been little research on multilayer tissue phantoms with tuneable blood flow, blood volume, and optical properties to simulate local changes in oxygenation at different depths. The aim of this study is to design, fabricate and characterize a complex dynamic phantom based on multilayer microfluidics with controllable blood flow, blood volume, and optical properties for testing NIRS instruments. We developed a phantom prototype with two microfluidic chips embedded at two depths inside a solid silicone phantom to mimic the vessels in the scalp and in the cortex. To simulate the oxygenation and perfusion of tissue, a solution with blood-like optical properties was sent into the microchannels by a pump with a programmable pressure controller. The pressure adjusted the volume of the microfluidic chips representing a distension of blood vessels. The optical changes in the superficial and deep layers were measured by a commercially available frequency domain NIRS instrument. The NIRS successfully detected the changes in light intensity elicited by the changes in the pressure input to the two layers. In conclusion, the microfluidics-based imaging phantom was successfully designed and fabricated and mimics brain functional activity. This technique has great potential for testing other optical devices, e.g., diffuse correlation spectroscopy, pulse oximetry, and optical coherence tomography.

Challenges and prospects for multi-chip microlens imprints on front-side illuminated SPAD imagers.

The overall sensitivity of frontside-illuminated, silicon single-photon avalanche diode (SPAD) arrays has often suffered from fill factor limitations. The fill factor loss can however be recovered by employing microlenses, whereby the challenges specific to SPAD arrays are represented by large pixel pitch (> 10 µm), low native fill factor (as low as ∼10%), and large size (up to 10 mm). In this work we report on the implementation of refractive microlenses by means of photoresist masters, used to fabricate molds for imprints of UV curable hybrid polymers deposited on SPAD arrays. Replications were successfully carried out for the first time, to the best of our knowledge, at wafer reticle level on different designs in the same technology and on single large SPAD arrays with very thin residual layers (∼10 µm), as needed for better efficiency at higher numerical aperture (NA > 0.25). In general, concentration factors within 15-20% of the simulation results were obtained for the smaller arrays (32×32 and 512×1), achieving for example an effective fill factor of 75.6-83.2% for a 28.5 µm pixel pitch with a native fill factor of 28%. A concentration factor up to 4.2 was measured on large 512×512 arrays with a pixel pitch of 16.38 µm and a native fill factor of 10.5%, whereas improved simulation tools could give a better estimate of the actual concentration factor. Spectral measurements were also carried out, resulting in good and uniform transmission in the visible and NIR.

Hybrid Convolutional Neural Network (hCNN) for Image Reconstruction in Near-Infrared Optical Tomography.

Near-infrared optical tomography (NIROT), a promising imaging modality for early detection of oxygenation in the brain of preterm infants, requires data acquisition at the tissue surface and thus an image reconstruction adaptable to cephalometric variations and surface topologies. Widely used model-based reconstruction methods come with the drawback of huge computational cost. Neural networks move this computational load to an offline training phase, allowing much faster reconstruction. Our aim is a data-driven volumetric image reconstruction that generalises well to different surfaces, increases reconstruction speed, localisation accuracy and image quality. We propose a hybrid convolutional neural network (hCNN) based on the well-known V-net architecture to learn inclusion localisation and absorption coefficients of heterogenous arbitrary shapes via a joint cost function. We achieved an average reconstruction time of 30.45 s, a time reduction of 89% and inclusion detection with an average Dice score of 0.61. The CNN is flexible to surface topologies and compares well in quantitative metrics with the traditional model-based (MB) approach and state-of-the-art neuronal networks for NIROT. The proposed hCNN was successfully trained, validated and tested on in-silico data, excels MB methods in localisation accuracy and provides a remarkable increase in reconstruction speed.

2.5 Hz sample rate time-domain near-infrared optical tomography based on SPAD-camera image tissue hemodynamics.

Time-domain near-infrared optical tomography (TD NIROT) techniques based on diffuse light were gaining performance over the last years. They are capable of imaging tissue at several centimeters depth and reveal clinically relevant information, such as tissue oxygen saturation. In this work, we present the very first in vivo results of our SPAD camera-based TD NIROT reflectance system with a temporal resolution of ∼116 ps. It provides 2800 time of flight source-detector pairs in a compact probe of only 6 cm in diameter. Additionally, we describe a 3-step reconstruction procedure that enables accurate recovery of structural information and of the optical properties. We demonstrate the system's performance firstly in reconstructing the 3D-structure of a heterogeneous tissue phantom with tissue-like scattering and absorption properties within a volume of 9 cm diameter and 5 cm thickness. Furthermore, we performed in vivo tomography of an index finger located within a homogeneous scattering medium. We employed a fast sampling rate of 2.5 Hz to detect changes in tissue oxygenation. Tomographic reconstructions were performed in true 3D, and without prior structural information, demonstrating the powerful capabilities of the system. This shows its potential for clinical applications.

Resolution and penetration depth of reflection-mode time-domain near infrared optical tomography using a ToF SPAD camera.

In a turbid medium such as biological tissue, near-infrared optical tomography (NIROT) can image the oxygenation, a highly relevant clinical parameter. To be an efficient diagnostic tool, NIROT has to have high spatial resolution and depth sensitivity, fast acquisition time, and be easy to use. Since many tissues cannot be penetrated by near-infrared light, such tissue needs to be measured in reflection mode, i.e., where light emission and detection components are placed on the same side. Thanks to the recent advance in single-photon avalanche diode (SPAD) array technology, we have developed a compact reflection-mode time-domain (TD) NIROT system with a large number of channels, which is expected to substantially increase the resolution and depth sensitivity of the oxygenation images. The aim was to test this experimentally for our SPAD camera-empowered TD NIROT system. Experiments with one and two inclusions, i.e., optically dense spheres of 5mm radius, immersed in turbid liquid were conducted. The inclusions were placed at depths from 10mm to 30mm and moved across the field-of-view. In the two-inclusion experiment, two identical spheres were placed at a lateral distance of 8mm. We also compared short exposure times of 1s, suitable for dynamic processes, with a long exposure of 100s. Additionally, we imaged complex geometries inside the turbid medium, which represented structural elements of a biological object. The quality of the reconstructed images was quantified by the root mean squared error (RMSE), peak signal-to-noise ratio (PSNR), and dice similarity. The two small spheres were successfully resolved up to a depth of 30mm. We demonstrated robust image reconstruction even at 1s exposure. Furthermore, the complex geometries were also successfully reconstructed. The results demonstrated a groundbreaking level of enhanced performance of the NIROT system based on a SPAD camera.

Localization of Deep Ischemia and Hemorrhage in Preterm Infants' Head with Near-Infrared Optical Tomography: A Numerical Case Study.

BACKGROUND AND AIM: Preterm infants have a high incidence of brain lesions that may lead to long-term disabilities. Early diagnosis of cerebral ischemia and hemorrhage may enable protection of the brain by prevention or neuroprotective treatment. Our recently developed time-domain near-infrared optical tomography (TD NIROT) system provides images to diagnose neonatal brain injury. Our aim is to study the image quality achievable from the TD NIROT signals perturbed by noise for two common cases: ischemia and hemorrhage. METHODS: We implemented simulations on a spherical model of diameter 60 mm representing a typical neonatal head where the absorption µa = 0.08 cm-1 and the reduced scattering µ's = 4.1 cm-1. Injury-mimicking spherical inclusions of various diameters (1 ~ 10 mm) were placed at depths of 10 ~ 20 mm in the ischemia case (2.5 × µa) and 14 ~ 30 mm for the hemorrhage case (50 × µa). TD data were generated from a large number of source-detector pairs, i.e., 208 detectors placed within a circle of diameter 40 mm on the surface surrounded by 18 sources. Up to 5% Gaussian noise was added in the simulations. 3D images were reconstructed with the modified Tikhonov minimization with the initial guess of a homogeneous phantom, and the images were evaluated by positional error and Dice similarity. RESULTS: The inclusions were localized correctly with low positional errors (<1 mm), and the segmented images share a high Dice similarity with the ground truth for both the ischemia and the hemorrhage case, even for tiny inclusions of 1 mm in deep tissue. The hemorrhage case with a high contrast tolerates a substantial level of noise even though the performance drops with higher noise as expected. CONCLUSIONS: The large amount of data provided by our novel TD NIROT system provides rich enough information for correctly locating hemorrhage and ischemia in the neonatal brain.

Dynamic time domain near-infrared optical tomography based on a SPAD camera.

In many clinical applications it is relevant to observe dynamic changes in oxygenation. Therefore the ability of dynamic imaging with time domain (TD) near-infrared optical tomography (NIROT) will be important. But fast imaging is a challenge. The data acquisition of our handheld TD NIROT system based on single photon avalanche diode (SPAD) camera and 11 light sources was consequently accelerated. We tested the system on a diffusive medium simulating tissue with a moving object embedded. With 3D image reconstruction, the moving object was correctly located using only 0.2 s exposure time per source.

Image reconstruction for novel time domain near infrared optical tomography: towards clinical applications.

Near infrared optical tomography (NIROT) is an emerging modality that enables imaging the oxygenation of tissue, which is a biomarker of tremendous clinical relevance. Measuring in reflectance is usually required when NIROT is applied in clinical scenarios. Single photon avalanche diode (SPAD) array technology provides a compact solution for time domain (TD) NIROT to gain huge temporal and spatial information. This makes it possible to image complex structures in tissue. The main aim of this paper is to validate the wavelength normalization method for our new TD NIROT experimentally by exposing it to a particularly difficult challenge: the recovery of two inclusions at different depths. The proposed reconstruction algorithm aims to tackle systematic errors and other artifacts with known wavelength-dependent relation. We validated the device and reconstruction method experimentally on a silicone phantom with two inclusions: one at depth of 10 mm and the other at 15 mm. Despite this tough challenge for reflectance NIROT, the system was able to localize both inclusions accurately.

Multimodal imaging combining time-domain near-infrared optical tomography and continuous-wave fluorescence molecular tomography.

Fluorescence molecular tomography (FMT) emerges as a powerful non-invasive imaging tool with the ability to resolve fluorescence signals from sources located deep in living tissues. Yet, the accuracy of FMT reconstruction depends on the deviation of the assumed optical properties from the actual values. In this work, we improved the accuracy of the initial optical properties required for FMT using a new-generation time-domain (TD) near-infrared optical tomography (NIROT) system, which effectively decouples scattering and absorption coefficients. We proposed a multimodal paradigm combining TD-NIROT and continuous-wave (CW) FMT. Both numerical simulation and experiments were performed on a heterogeneous phantom containing a fluorescent inclusion. The results demonstrate significant improvement in the FMT reconstruction by taking the NIROT-derived optical properties as prior information. The multimodal method is attractive for preclinical studies and tumor diagnostics since both functional and molecular information can be obtained.

Validation and Comparison of Monte Carlo and Finite Element Method in Forward Modeling for Near Infrared Optical Tomography.

Near infrared optical tomography (NIROT) is a non-invasive imaging technique to provide physiological information e.g. the oxygenation of tissue. For image reconstruction in clinical and preclinical scenarios, models to accurately describe light propagation are needed. This work aims to assess the accuracy and efficiency of different models, which paves the way for an optimal design of model-based image reconstruction algorithms in NIROT for realistic tissue geometries and heterogeneities. Two popular simulators were evaluated: the Monte Carlo (MC) method based MCX and the finite element method (FEM) based Toast++. We compared simulated results with experimental data measured on a homogeneous silicone phantom with well-calibrated parameters. The laser light was focused on the center of the phantom surface and images were captured by a CCD camera in both reflection and transmission modes. For transmittance measurements, the two models showed good agreement. Both achieve a cosine similarity of ~99%. In contrast, for reflectance measurements, FEM results deviated more from the measured values than MC, yielding similarity values of 86% and 94%, respectively. This study recommends the use of MC for NIROT in reflection mode and both MC and FEM yield excellent results for transmission mode.

Absorption spectra of early stool from preterm infants need to be considered in abdominal NIRS oximetry.

Necrotizing enterocolitis (NEC) is the most common gastrointestinal emergency of the preterm infant. Low abdominal tissue oxygen saturation (StO2) measured by near-infrared spectroscopy (NIRS) oximetry may be an early sign of NEC relevant for treating or even preventing NEC. However, current commercial NIRS oximeters provide inaccurate StO2 readings because they neglect stool as an abdominal absorber. To tackle this problem, we determined the optical properties of faeces of preterm infants to enable a correct abdominal StO2 measurement. In 25 preterm born infants (median age 31 0/7 ± 2 1/7 weeks, weight 1478 ± 511 g), we measured their first five stool probes with a VIS/NIR spectrometer and calculated the optical properties using the Inverse Adding Doubling (IAD) method. We obtained two absorption spectra representing meconium and transitional stool. Probabilistic cluster analysis correctly classified 96 out of 107 stool probes. The faeces spectra need to be considered to enable correct abdominal StO2 measurements with NIRS oximetry.

Discrimination of Complex Activation Patterns in Near Infrared Optical Tomography with Artificial Neural Networks.

Near-infrared optical tomography (NIROT) has great promise for many clinical problems. Here we focus on the study of brain function. During NIROT image reconstruction of brain activity, an inverse problem has to be solved that is sensitive to small superficial perturbations on the head such as e.g. birthmarks on the skin and hair. To consider these perturbations, standard physical modeling is unpractical, since it requires the implementation of detailed information that is generally unavailable. The aim here was to test whether artificial neural networks (ANN) are able to handle such perturbations and thus detect brain activity correctly. For simplicity, we created a virtual test model, where we simulated a pattern of activated and resting brain regions, which was covered by skin features like hair or melanin. We compared the performance of this ANN approach with that of an inverse problem based on a Monte Carlo (MC) model for light propagation. We conclude that ANNs tolerate substantially higher levels of skin perturbations than MC models and consequently are more suitable for detecting brain activity.

A New Method Based on Virtual Fluence Detectors and Software Toolbox for Handheld Spectral Optoacoustic Tomography.

A minimal setup for optoacoustic (OA) imaging requires an ultrasound probe and a pulsed laser. Such a system is capable of imaging small blood vessels and is sensitive to variations in their oxygen saturation. However, absolute oxygenation values cannot be obtained without a proper correction for the varying light fluence resulting from the optical attenuation in the surrounding tissue. Other techniques, such as near-infrared optical tomography (NIROT) can be employed to assist OA imaging for fluence compensation. In this paper, we propose using blood vessels as virtual fluence detectors (VD), which serve as light detectors for NIROT image reconstructions. By avoiding the use of real photon detectors, a simpler system could be implemented in a hand-held device comparable in size with conventional ultrasound probes. Even for a low number of VDs it provides increased informational value which, in combination with a large number of light sources, results in precise reconstructions. We define a tomographic inverse problem based on ratios of OA signals measured at several wavelengths where optical properties of VDs, tumor and normal tissue can be reconstructed simultaneously. The use of ratio data effectively removes light source skin coupling errors for the case of emission in a single point, which is required for clinical applications. We have defined the mathematical structure of an inverse problem where chromophore concentrations for normal, tumor and embedded VDs are obtained simultaneously from this ratio data. To test the performance of our approach we show an image reconstruction on a virtual phantom with an embedded tumor in the vicinity of eight blood vessels. We conclude that this limited number of VDs, located in areas of maximum sensitivity result in high quality reconstructions. For the simplest case of a single blood vessel located in a homogeneous tissue, we present a graphical user interface based toolbox for conducting virtual experiments. The toolbox can be used to assist in the design and optimization of suitable hardware for different applications, among which imaging tumor oxygenation and ischemic lesions in the brain of preterm infants are of great clinical value.

Optical properties of mice's stool in 550 to 1000 nm wavelength range.

The aim of this work was to measure optical properties of stool of mice to provide this relevant wavelength-dependent behavior for optical imaging modalities such as fluorescent molecular tomography and near-infrared optical tomography. BALB/c nude female mice were studied and optical properties of the stool were determined by employing the inverse adding-doubling approach. The animals were kept on chlorophyll-free diet. Nine stool samples were measured. The wavelength-dependent behavior of absorption and scattering in 550 to 1000 nm range is presented. The reduced scattering spectrum is fitted to the Mie scattering approximation in the near-infrared (NIR) wavelength range and to the Mie + Rayleigh approximation in visible/NIR range with the fitting coefficients presented. The study revealed that the absorption spectrum of stool can lead to crosstalk with the spectrum of hemoglobin in the NIR range.

Development and Validation of a Sensor Prototype for Near-Infrared Imaging of the Newborn Brain.

Imaging brain oxygenation is crucial for preventing brain lesions in preterm infants. Our aim is to build and validate a near-infrared optical tomography (NIROT) sensor for the head of neonates. This sensor, combined with an optoacoustic device, will enable quantitative monitoring of the structural and functional information of the brain. Since the head of preterm infants is small and fragile great care must be taken to produce a comfortable and compact device in which a sufficient number of light sources and detectors can be implemented. Here we demonstrate our first prototype. Heterogeneous silicone phantoms were produced to validate the prototype's data acquisition, data processing, and image reconstruction. Reconstructed optical properties agree well with the target values. The mechanical performance of the new NIROT sensor prototype confirms its suitability for the clinical application.

A New Method Based on Graphics Processing Units for Fast Near-Infrared Optical Tomography.

The accuracy of images obtained by Diffuse Optical Tomography (DOT) could be substantially increased by the newly developed time resolved (TR) cameras. These devices result in unprecedented data volumes, which present a challenge to conventional image reconstruction techniques. In addition, many clinical applications require taking photons in air regions like the trachea into account, where the diffusion model fails. Image reconstruction techniques based on photon tracking are mandatory in those cases but have not been implemented so far due to computing demands. We aimed at designing an inversion algorithm which could be implemented on commercial graphics processing units (GPUs) by making use of information obtained with other imaging modalities. The method requires a segmented volume and an approximately uniform value for the reduced scattering coefficient in the volume under study. The complex photon path is reduced to a small number of partial path lengths within each segment resulting in drastically reduced memory usage and computation time. Our approach takes advantage of wavelength normalized data which renders it robust against instrumental biases and skin irregularities which is critical for realistic clinical applications. The accuracy of this method has been assessed with both simulated and experimental inhomogeneous phantoms showing good agreement with target values. The simulation study analyzed a phantom containing a tumor next to an air region. For the experimental test, a segmented cuboid phantom was illuminated by a supercontinuum laser and data were gathered by a state of the art TR camera. Reconstructions were obtained on a GPU-installed computer in less than 2 h. To our knowledge, it is the first time Monte Carlo methods have been successfully used for DOT based on TR cameras. This opens the door to applications such as accurate measurements of oxygenation in neck tumors where the presence of air regions is a problem for conventional approaches.