1. Introduction

Over the last two decades, multi-modal instrumentation combining near-infrared diffuse optical spectroscopy and ultrasound (US) began to emerge in clinical research and applications [1–4]. The bulk of this work focused on oncological applications where the optical methods are hypothesized to provide complementary physiological information to ultrasound and other modalities. US is utilized as a convenient and clinically accepted modality to guide the optical measurements by identifying the right region of interest, and by providing information about the geometry of the underlying tissue, which, in turn, is used to improve the accuracy and the precision of the optical data analysis [5–11].

Diffuse optical technologies utilize near-infrared light ($\sim 650-950\;nm$) [12,13] and have gathered interest for testing for a broad range of potential clinical applications [14–21]. They are relatively unique since they can probe deep ($>$1 cm) tissues in a non-invasive, safe manner to recover important quantitative information about tissue hemodynamics, components and structural features. These parameters are speculated to be important biomarkers of both healthy functioning tissue and its pathology [12].

Near-infrared time resolved spectroscopy (TRS) [22–24] utilizes short laser pulses ($\sim 100\;ps$) at multiple wavelengths and measures their delay and broadening after having traveled through the tissue. It is capable of determining the absolute values of tissue optical properties such as the wavelength dependent absorption and reduced scattering coefficients, which are then used to retrieve physiological quantitative information. Of particular interest are the concentration of oxy- and deoxy-hemoglobin species, which are used to derive the microvascular blood oxygenation and blood volume. Other important components often measured by TRS are collagen, water and lipid concentrations [25,26]. In addition, the measurement of the scattering coefficient can provide important information about tissue cellular structure [27,28].

Diffuse correlation spectroscopy (DCS) [12,29,30], on the other hand, uses a continuous wave coherent laser source and exploits the decay of the speckle intensity auto-correlation function due to light scattering by moving particles (i.e. red blood cells) to measure the bulk microvascular blood flow of the probed tissue volume.

Separately, DCS and TRS only offer limited information about tissue physiology, but, the simultaneous combination of the information provided by TRS and DCS allows to retrieve complementary information about hemodynamics, oxygen metabolism, composition and structure of the measured tissue, enhancing the sensitivity to those biomarkers that important to discriminate between healthy and pathological conditions. This has been successfully exploited in several clinical research studies, from adult and infant brain monitoring to cancer diagnosis and therapy monitoring [31–38].

In this context, a consortium titled “Laser and Ultrasound Co-Analyzer for thyroid nodules” (LUCA) has been formed, funded by the European Commission (see Appendix) [39]. The overall goal of the LUCA project is to develop a multi-modal point-of-care device, combining time resolved spectroscopy, diffuse correlation spectroscopy and medical ultrasound imaging in a single platform with the aim of improving the screening of thyroid nodules for cancer.

Thyroid nodules are a common pathology having a prevalence of palpable nodules of around 5% in women and 1% in men, and, when neck ultrasound is used for screening, the prevalence increases to 19-76% [40,41]. The current established diagnosis protocols for assessing the malignancy of a nodule envisage the thyroid examination with ultrasound and ultrasound Doppler, followed by a fine needle aspiration biopsy (FNAB). Unfortunately, these diagnosis procedures have limited effectiveness, lacking in specificity and sensitivity (ranging from 10-87% in sensitivity to 58-96% in specificity depending on the type of the malignant nodules), and lead to a large number of non-diagnostic and/or false positives, resulting in unnecessary surgeries [42–46].

LUCA platform will try to address the poor sensitivity and specificity of current thyroid cancer screening protocols by providing at the same time information about anatomy (through the medical US imaging), tissue hemodynamics (microvascular blood flow, blood oxygenation), and chemical constitution (collagen, water and lipid concentrations, and scattering spectrum) of the tissue examined [10,25,47].

In the following sections, we present the detailed description of the LUCA device, considering all the separate modules and their integration in a single device, together with the tests performed in tissue mimicking phantoms and in vivo.

2. LUCA device and methods

LUCA aim is to be integrated inside the current thyroid nodule screening work-flow, by providing additional physiological information simultaneously to the clinical ultrasound examination performed by the medical doctor. For this purpose, all the different technologies of LUCA have been embedded in a single device. A unique multi-modal probe, including both the US transducer and the optical fibers, has been developed to allow the simultaneous acquisition of optical and US signals. For the same purpose, the LUCA prototype has been built in such a way that the measurement protocol can be managed by the medical doctor through the US touch-screen interface, capable of visualizing and storing the results of optical measurements together with the real time ultrasound images. An additional software and a separate monitor have been added to control and set advanced parameters related to the optical subsystems as needed, which is particularly important during the first clinical measurement campaigns to optimize the optical parameters and measurement protocols.

A photo of the LUCA device is shown in Fig. 1(a), together with the LUCA probe (b), and a diagram of the different modules of the device (c). The modules are: a time resolved spectroscopy module (TRSM), a diffuse correlation spectroscopy module (DCSM), an ultrasound module (USM), a multi-modal optical-ultrasound probe, a main control module (MM) responsible of the control and communication with all the other modules, and a post processing evaluation module based on NIRFAST software (NIRFAST evaluation module, NEM) [48]. In addition, the LUCA device includes an uninterruptible power supply, an emergency shut-down button, and a medical grade isolation transformer.

 

Fig. 1. (a) Photo of the LUCA device, with all the modules. (b) Photo of the LUCA multi-modal optical-ultrasound probe, with a sketch of the probe nose showing the current geometrical disposition of the fiber tips: $x-y$ DCS source fibers; $z$ TRS source fiber, $m-n$ DCS detection fibers, $p-q$ TRS detection fibers. (c) Block diagram of the LUCA device modules and communication.

Download Full Size | PDF

2.1 Time resolved spectroscopy module (TRSM)

The TRSM - described as a stand-alone system in a separate publication [49] - has been developed by LUCA partner Politecnico di Milano. It is hosted in a 19” case ($48\times 38\times 20\;cm^{3}$) and has eight gain-switched laser diode sources, with wavelengths in the range $635-1050\;nm$, that are driven at a repetition rate of $40\;MHz$ by a custom-made electronic control board with low-jitter logic. Such laser sources feature a temporal jitter of about $140\;ps$ (full width half maximum, FWHM), with output powers greater than $1\;mW$.

Laser outputs are coupled with standard graded index (GRIN) $100/140\;\mu m$ core/cladding diameter fibers, which, in turn, are the input of a $9\times 1$ optical switch (Piezosystem Jena, Germany), which selects the wavelength that is injected sequentially into the sample through standard GRIN fibers.

Diffuse photons are recollected at two source-detector distances by using two compact silicon photomultipliers (SiPM) detectors, which show state-of-the-art performances compared to bulky, expensive and fragile traditional systems based on photomultiplier tubes. Notch filters (Semrock, U.S.A.) have been added in front of the detectors to filter out the DCS source light at $785\;nm$, allowing simultaneous DCS and TRS acquisition.

Two time-to-digital converters (TDC), developed in application specific integrated circuits (ASIC) and integrated onto a field programmable gate array (FPGA) board, reconstruct the distribution of time-of-flight of photons (DTOF) up to count-rates of $3\;Mcounts/s$ with a timing resolution of about $10\;ps$.

2.2 Diffuse correlation spectroscopy module (DCSM)

The DCSM is a custom DCS device built at ICFO. It is mounted in an independent 19” case, with a independent medical grade modular power supply.

The DCSM source is custom made by LUCA partner HemoPhotonics S. L., and consists of a fiber coupled single longitudinal mode continuous wave laser emitting at $785\;nm$, with a typical operation power of $70\;mW$. An optical clean up filter at $785\; nm$ (Semrock, U.S.A.) is inserted in the laser module to remove the tails of the laser spectral emission. The emitted light is split by a fiber beam splitter into two equal intensity branches, allowing the DCSM having two laser outputs coherent with each other, that is, with the relative phase constant over a time longer than the acquisition time of every correlation curve. The intensity of each output is tunable by two fiber optical attenuators. For the in vivo measurements, each fiber’s power output has been limited to $27\;mW$ to respect the safety limits established by American National Standards Institute (ANSI) and by the International Organization for Standardization (ISO, IEC 60825-1:2014, European Union normative EN 60825-1:2014) (for determining the maximum permissible exposure MPE, the irradiance has been calculated considering a limiting circular aperture of $3.5\;mm$ diameter) [50–52].

The DCSM is characterized by sixteen detection channels that are utilized to increase the signal-to-noise ratio in cases of low detected photon count-rates [53,54]. The intensity autocorrelation functions are calculated real time by a custom developed low-cost multi-channel (16 channels), multi-tau, fast ($100\;Hz$) hardware auto-correlator made by four stackable and synchronized field programmable gated arrays (FPGA) of four channels each (HemoPhotonics, Castelldefels, Spain).

2.3 Ultrasound module (USM)

The LUCA ultrasound module is based on a commercial ultrasound device, EXAPad, produced by IMV Imaging, France [55]. EXAPad has a touchscreen interface, and allows several functionalities such as B, B+B, B+M and M imaging, color Doppler, pulse wave and continuous wave Doppler modes.

The standard commercial EXAPad US module software has been modified by the manufacturer in order to deal with the communication with the main module and to manage the LUCA measurement protocol. In addition to the standard US functionalities, it includes a window tab dedicated to LUCA. Through this tab, the user connects the EXAPad module to the main module (via TCP/IP connection), in order to handle the data and image transfer to the optical main module, and to visualize and store the results of the optical measurements together with the US images acquired simultaneously. Moreover, it includes interfaces to start the optical measurement in “preview” mode, in order to check the optical data quality, and to run the optical measurement in “acquisition” mode. Lastly, the EXAPad module includes a section to review the images acquired and start the post-processing with the NIRFAST evaluation module (see below).

2.4 Multi-modal ultrasound-optical probe

The custom LUCA multi-modal optical-ultrasound probe has been developed by LUCA partner Vermon, France. It includes a standard US transducer commonly utilized for thyroid ultrasound ($192$ elements in the transducer, average sensitivity of $168\;mV$, a $70\%$ relative bandwidth and a $10.6\;MHz$ center frequency), and a chassis that allows the insertion of the optical fibers. The optical fiber tips are fixed to two replaceable fiber holders at the nose of the probe, one for the source fibers, and one for the detection fibers, placed next to the two long sides of the transducer (see Fig. 1(b)).

The design, together with the replaceable holders, allows flexibility for defining the geometry of the fiber tip positions, permitting the mounting of different holders with different geometries on the probe nose. The current geometry of the LUCA probe holders is reported in Fig. 1(b), together with a photo of the probe. Both DCS and TRS fiber configurations have two source-detector separations, the short distance, more sensitive to superficial layer properties, of $19\;mm$ and the long, more sensitive to deeper tissue properties, of $25\;mm$. This configuration allows to simultaneously perform TRS and DCS measurements without crosstalk between the two technologies by reducing the effect of the DCS sources on TRS detectors and vice versa (see section 3.2). Furthermore, it is built around the standard US probe mechanics.

The TRS source fiber consists of a $100/140\;\mu m$ core/cladding diameter graded index fiber, and the two detection patchcords consist in two fiber bundles of $168$ graded index $100/140\;\mu m$ core/cladding diameter fibers (long source-detector separation) and $213$ graded index $62.5/125\;\mu m$ core/cladding diameter fibers (short source-detector separation).

The DCS source at the probe nose consists of two $400\;\mu m$ core fibers with ferrule termination placed at a reciprocal distance (center-to-center) of $4.4\;mm$ in a symmetrical position respect to the detection fibers axis (see Fig. 1(b)). This particular configuration allows to roughly double the photon intensity detected by the DCS detectors, in compliance with the laser safety limits for tissue illumination, with respect to the standard configuration with a single DCS source fiber. The DCS detection fibers consist of a single mode fiber (for the wavelength of $785\;nm$) of $4.4/125\;\mu m$ core/cladding diameter for the short source-detector separation, and in a bundle of fifteen single mode fibers ($4.4/125\;\mu m$ core/cladding diameter) for the long source-detector separation.

In addition, the LUCA probe includes a capacitive skin contact sensor, to detect bad coupling between the optical fiber tips at the probe nose and the skin, and an extra laser enable push button, which is used to enable the laser emission only when the probe is placed in a safe position on the tissue.

2.5 Post-processing, NIRFAST evaluation module (NEM)

The post-processing module consists of a custom assembled computer, with a graphics processing unit (GPU) utilizing CUDA (Compute Unified Device Architecture, Nvidia, U.S.A.) technology supporting massive-parallelization in code execution [56,57]. Tissue parameter recovery is based on established NIRFAST software [48] package expanded with curve-fitting of TRS and DCS data using finite element method modelling. The NEM communicates through a network (TCP/IP) connection with the main module to receive the multi-modal data (the US images together with the optical raw data associated) and to return results (thyroid hemodynamic parameters and chromophore concentrations). The NIRFAST evaluation module is mounted in an independent 19” case, powered by a medical grade power supply. The post-processing system has been tested within the LUCA device environment on phantom and in-vivo data [58], and, as a sufficient number of subjects is recruited in the LUCA clinical campaign, the finite-elements solver of NIRFAST as well as its inversion algorithms will be utilized to model the curved and heterogeneous structures such as the trachea, carotid arteries as well as the thyroid tissue and the overlying muscles based on the ultrasound images. This will be the subject of a separate publication.

2.6 Main module (MM)

The main module is the principal control unit for the operation of the optical modules. It communicates with optical modules via USB connections, provides the real-time evaluation of optical data, and communicates with the ultrasound module via TCP/IP connection. We note that the user interacts with the LUCA device primarily through the USM. The operator can set optical evaluation parameters, supervise optical data quality and optical data acquisition in real time within the USM application, which requires such information from the MM. Additionally, the MM handles the requests for NEM post-processing via TCP/IP connection and stores optical and US data sets.

The main functional units of the MM are: i) an industrial PC equipped with the required standard communication ports for USB and TCP/IP communication (ethernet port); ii) a custom electronic board for overall control including control of individual components of other modules. This board is equipped with a microcontroller and has been developed for the LUCA device for all lower level control lines (TTL) in order to enable/disable single module components, trigger options and safety features for the DCSM, TRSM as well as the LUCA probe. It includes several units of discrete safety electronics (for interlocks) and isolated readout/supply channels (for electrical safety) for probe electronics operation. The schematics of the Main Module connections with the other LUCA modules is reported in Fig. 1(c).

In order to follow the modular concept of the LUCA device, the industrial PC and the microcontroller board are combined in an independent 19” case.

3. Device characterization

In this section we present the tests performed for the characterization of the LUCA device, mainly focusing on the optical modules. US module and the US probe, as well as any customization that was carried out, were tested by the manufacturers using standard procedures. Every module has been thoroughly tested individually and the whole LUCA device has been tested as a unique integrated system. Particular care has been dedicated to test the interference and cross-talk between all the different technologies working together in LUCA.

Here, we report DCS and TRS phantom tests, and in vivo measurements such as repeatability in thyroid and arm cuff occlusion demonstration tests.

3.1 Analysis method for TRS and DCS acquisition

The reduced scattering ($\mu _{s}'$) and absorption ($\mu _{a}$) coefficients related to all the different wavelengths are retrieved from TRS acquisition by non-linear fitting of the solution of the diffusion equation for a semi-infinite homogeneous medium [59–61], convoluted with the instrument response function (IRF) to the DTOF. The IRF is acquired separately for the long and short separation detection channels by placing in contact the tips of the source and detection fibers.

The effective Brownian diffusion coefficient $D_{B}$ (for phantom measurements) and the blood flow index $BFi$ (for in vivo measurements) have been retrieved by fitting of the solution of the electric field correlation diffusion equation for a semi-infinite homogeneous medium to the measured intensity auto-correlation functions [12]. The reduced scattering and absorption coefficients have first been obtained through TRS curve fitting, and then used as fixed parameters in the DCS curve fitting.

For retrieving the oxy and deoxy-hemoglobin changes ($\Delta HbO_{2}$ and $\Delta Hb$) in the in vivo tests (see section 3.5), we have assumed linearity between absorption coefficients $\mu _{a} (\lambda )$ and the chromophore concentrations $c_{i}$, through the relation $\mu _{a}(\lambda ) = \sum _{i}\epsilon _{i}(\lambda )c_{i}$, where $\epsilon _{i}$ is the $i$-th chromophore’s extintion coefficient. The only parameters fitted were oxy and deoxy-hemoglobin concentrations, while water concentration has been kept fixed at $70\%$ of the total chromophore concentration.

3.2 Integration of all the technologies

The optical and electronic interference between the two different diffuse optical technologies (TRS and DCS) and ultrasound has been evaluated during the design and the construction phases of the project. Additional precautions were taken whenever needed such as the insertion of notch and clean-up laser filters in TRSM and DCSM, and smart geometrical fiber disposition in the probe.

The effects that can reduce the quality of the acquisition are linked to the possible interference between US waves and optical acquisitions, between TRS sources and the DCS detectors, and between DCS sources and TRS detectors. In addition, we have tested the effect of the presence of the US coupling gel, in between the tissue and the probe, on the optical measurements quality. These last tests were the subject of a dedicated publication [62].

The cross-talk between US and optical acquisitions has been evaluated by performing measurements with the different modules alternatively turned on and off, and comparing these results to simultaneous acquisition. The results obtained demonstrates that TRS and DCS do not affect the quality of the US images, neither the presence of US waves affect the quality of TRS and DCS acquisitions (data not shown).

The cross-talk between TRS and DCS has been evaluated by measuring the background counts of DCS with TRS laser ON and OFF, in phantom and in thyroid (and, vice versa, by measuring the background counts of TRS with DCS laser ON and OFF). The phantom used consisted of a solid phantom based on epoxy resin with black toner as absorber and $TiO_{2}$ powder as scatterer [62,63], with reduced scattering coefficient of $10\;cm^{-1}$ and absorption coefficient of $0.1\;cm^{-1}$ at a wavelength of $660\;nm$. This situation is considered an extreme and unrealistic situation in the thyroid region, which is characterized by different optical properties (above all a much higher absorption, that reduces the effect of DCS laser on TRS detectors and vice versa) [10,47].

In Fig. 2(a) and (b), we have reported the effect of DCS laser on TRS acquisitions. While when measuring on the thyroid, the measured cross-talk between DCS and TRS is practically null (no background count-rate changes), we have registered an increase of approximately $100\%$ of the TRS detector background when DCS source is turned on (background counts increases from $\sim 50\;kcounts/s$ to $\sim 100\;kcounts/s$, after turning on DCS laser). As reported in Fig. 2(a) and (b), this increase does not affect the retrieval of the absorption and reduced scattering coefficients, since the TRS signal count-rate (roughly $1\;Mcounts/s$) assure in any case a good signal-to-noise-ratio. Lastly, with regard to DCS, the measured background changes ($<0.3\;kcounts/s$) are negligible considering the common DCS count-rates in tissue and thyroid in particular ($>10\;kcounts/s$) [10,12] and do not affect the acquisition. The minor differences (an example is reported in Fig. 2(c)) are well below the variability associated to tissue physiological changes during the measurement.

 

Fig. 2. Cross-talk between TRS and DCS. (a) Absorption coefficient measured by TRS in thyroid and phantom with DCS ON and OFF. (b) Reduced scattering coefficient measured by TRS in thyroid and phantom with DCS ON and OFF. (c) Blood flow index measured by DCS in thyroid with TRS ON and OFF

Download Full Size | PDF

3.3 TRS phantom characterization

The time resolved spectroscopy module has been characterized and validated as an independent device through several phantom tests associated to the BIP, MEDPHOT and nEUROPt standardization protocols [63–66]. Preliminary results, related to the TRS as an independent stand-alone system, have already been published in a separate, dedicated publication [49]. Here we report a summary of the performance assessment standardization protocols (i.e. BIP, MEDPHOT and nEUROPt) implemented for the LUCA TRS integrated in the complete LUCA system. In order to not make this section heavier for the reader, the full set of results and detailed description of the tests performed have been reported in the Supplement 1.

BIP protocol characterizes the main features (source and detection components) of the TRS part of the LUCA device [64]. The optical power delivered to the sample (fully open attenuators) ranges from $0.62\;mW$ (at $730\;nm$) up to $2.90\;mW$ (at $830\;nm$). The responsivity of the detection chain is reported in the Supplement 1. Time to digital converter differential non linearity ($DNL\;\%$) is around $15\;\%$ for both the detection channels. Finally, the afterpulsing ratio, $AR\;\%$, was evaluated. We have estimated an afterpulsing ratio of few percent (from $0.6\;\%$ at 635 nm up to $8.5\;\%$ at 1040 nm), with a dependence on the wavelength probably due to the long decaying tail of the SiPM detectors. While this point will require further investigations, the small AR does not affect at all the reconstruction of the TRS data. Concluding, the TRS part of the device features an overall temporal resolution (FWHM of the instrument response function, IRF) ranging from $106$ up to $193\;ps$ depending on the wavelength, and it is stable over several hours ($>6$ hours) within an error $1\;\%$ of the average after thirty minutes from switching on of the device.

MEDPHOT protocol assesses the capability of the device to retrieve homogeneous optical properties of the tissue/sample under investigation. A $32$ phantoms kit has been used to assess the linearity, accuracy and the reproducibility of the device [63]. From the tests performed we saw, as expected, a linear scaling of measured absorption and reduced scattering coefficients with respect to the nominal values of the phantoms used. The accuracy measured in retrieving the nominal absorption and reduced scattering coefficients is better than $20\;\%$ for the entire range of wavelengths.

The precision of the measured parameters $\mu _{s}'$ and $\mu _{a}$, defined by the coefficient of variation $CV=100\cdot \sigma (x)/\langle x\rangle$ where $x$ is the measured parameter, $\langle x\rangle$ its average value, and $\sigma (x)$ the standard deviation, is found to be better $1\;\%$ when the measurements are acquired at a count-rate higher than $200\;kcounts/s$ at all the wavelengths. Moreover, we found $CV$ for both absorption and scattering, to be fairly in accordance with the ideal dependence $N^{-0.5}$, where $N$ is the count-rate, which represents Poisson’s noise contribution to the signal. Finally, the reproducibility of the retrieval of optical properties, obtained by repeating the measurement on the same phantom in different days, resulted better than $3\;\%$ for the absorption and $4\;\%$ for the scattering coefficient for both detection channels and at all the wavelengths. Overall, these values are within the requirements for successful in vivo use.

nEUROPt protocol evaluates the capability of a time resolved system to detect, localize, and quantify absorption changes within tissues. In this framework, phantoms with small black inclusions are used to mimic absorption changes [65,66]. As expected the maximum of contrast is obtained when the inclusion is below the probe and the value increases at later time-gates to finally drop due to the presence of the long time-constant tail of the SiPMs [67].

3.4 DCS phantom characterization

The DCS module has been thoroughly tested in the laboratory setting to assess its performance through liquid phantom experiments. A photo of the experimental setup is reported in Fig. 3(a). The stability of the system has been checked by measuring the Brownian diffusion coefficient $D_{B}$ of a liquid phantom, over long time experiment (17 hours). The liquid phantom consists of a suspension of Lipofundin20% (B. Braun Melsungen AG, DE) in water, prepared following the recipe reported in Ref. [68] in order to have a reduced scattering coefficient of $5\;cm^{-1}$ and water absorption. The measured $D_{B}$ over time, related to one acquisition channel, at a photon count-rate of $200\;kcounts/s$, and average time $1\;s$, is reported in Fig. 3(b). The relative standard deviation over 17 hours measurement resulted $<3\%$. In Fig. 3(c) we report the standard deviation of the reduced autocorrelation function ($g_{2}-1$), for all the acquisition channels over the whole measurement time. The results reported resulted in line with the state of the art diffuse correlation spectroscopy devices [35,69,70].

 

Fig. 3. DCS module phantom demonstration tests. (a) Photo of the experimental setup, with the LUCA probe fixed through a mechanical holder in vertical position over the liquid phantom box. (b) Stability of $D_{B}$ over long time test (approx. 17 hours), data reported for one detection channel. (c) Standard deviation of the reduced autocorrelation function ($g_{2}-1$), for all the 16 acquisition channels, over 17-hours stability test. (d) Measured $D_{B}$ for phantom with different glycerol concentrations, long source-detector separation. (e) Ratio of the measured $D_{B}$ for phantom with different glycerol concentration (i. e. different viscosity) and comparison with theory, long source-detector separation. (f) Precision ($CV$) of the measured $D_{B}$ for phantom with different glycerol concentration (i. e. different viscosity), long source-detector separation. Different colors represent solutions with different absorption coefficient (blue $\mu _{a}=0.1\;cm^{-1}$, red $\mu _{a}=0.2\;cm^{-1}$, green $\mu _{a}=0.3\;cm^{-1}$), while the displacement with respect to the vertical axis represents different reduced scattering coefficients (left displacement: $\mu _{s}'=5\;cm^{-1}$, center: $\mu _{s}'=10\;cm^{-1}$, right displacement: $\mu _{s}'=15\;cm^{-1}$). (g), (h), and (i) same as (d), (e) and (f) but for short source-detector separation channel.

Download Full Size | PDF

The capability of the LUCA DCS module to detect $D_{B}$ changes has been proven by preparing a phantom with different viscosities obtained by adding glycerol in the suspension on Lipofundin20% in water [68], for different nominal values of reduced scattering coefficient ($5$, $10$ and $15\;cm^{-1}$) and absorption coefficient ($0.1$, $0.2$ and $0.3\;cm^{-1}$).

The particle Brownian diffusion coefficients related to the different phantoms have been measured and reported in Fig. 3(d) for the long source-detector separation channels, and (g) for the short one. Here, the values of the reduced scattering and absorption coefficients used in the DCS fitting procedure are measured simultaneously with LUCA TRS acquisition. This figure highlights the fact that, for increasing glycerol concentration, the measured particle Brownian diffusion coefficient decreases, as expected. Moreover, for highly absorbing and scattering phantoms, the Brownian diffusion coefficient measured results overestimated with respect to what is measured in phantoms with lower absorption and scattering. This behavior has been already reported in Ref. [69], and depends on the difficulty of determining the $\beta -factor$ in regime of low detected count-rates.

The capability of the LUCA DCS module to correctly measure changes in $D_{B}$ results evident in Fig. 3(e) for the long source-detector separation channels - (h) for the short one - where we report the ratios of the measured $D_{B}$ with the $D_{B}$ of the $0\%$ glycerol correspondent phantom, together with the expectation of the theory (i.e. $D_{B1}/D_{B2}=\eta _{2}/\eta _{1}$, where $\eta$ is the viscosity of the solution [68,71]). We note that the accuracy of the recovered Brownian diffusion coefficient from these typical phantoms can be reliably assessed and is an ongoing topic of research [68,72–74].

Finally, in Fig. 3(f) and (i) the coefficient of variation $CV$ of the retrieved $D_{B}$ (obtained by averaging ten subsequent acquisitions on the same phantom) is reported for all the phantoms measured, for long and short source-detector separation respectively. The $CV$ determines the precision of the $D_{B}$ measurements, and it is mainly affected by the detected photon count-rate, that is lower for high scattering and absorption values.

3.5 In vivo characterization

3.5.1 Variability on thyroid

To assess the precision of the LUCA device in measuring the optical and hemodynamic properties of the thyroid tissue, repeatability tests have been performed. The same thyroid lobe of the same healthy volunteer has been measured in four different days in two consecutive weeks. Each day, we have performed five independent measurements of the same lobe, by removing and replacing the LUCA probe on the same position of the volunteer neck. By looking at the real time US images of neck, the operator was able to find the same position after every probe replacement. This operation requires approximately one minute of time. The LUCA probe has been fixed to volunteer neck in a stable position through a mechanical arm, that reduces the artefacts due to probe movements along the duration of each single acquisition (approximately $100\;s$). In addition, artefacts in the signal clearly due to subject movements, have been detected and removed when processing the data. In Fig. 4(a) we report images of the experimental configuration adopted for the in vivo repeatability tests. In Fig. 4 (b) we show an US image of the thyroid of the healthy voluntary subject, acquired simultaneously to the optical acquisition.

 

Fig. 4. In vivo tests. (a) Set up for thyroid repeatability measurements. (b) US image acquired simultaneously with optical data: 1- thyroid right lobe; 2- carotid; 3- sternocleidomastoid muscle; 4- skin; 5- trachea. (c) DCS repeatability results in the right thyroid lobe (5 repetitions per day, 4 days of measurements). (d) TRS repeatability results in the right thyroid lobe, absorption coefficient (top) and reduced scattering coefficient (bottom).

Download Full Size | PDF

The main results of the thyroid repeatability tests are summarized in Table 1 for the DCS and Tables 2, 3 and 4 for the TRS, and reported graphically in Fig. 4(c) (DCS) and (d) (TRS). Overall, these preliminary results in thyroid are in line with what was reported in Ref. [10]. A deeper quantitative comparison with previous literature in vivo results falls beyond the scope of this work, and will be available as the LUCA clinical research campaign, currently ongoing, reaches a sufficient number of recruited subjects. Furthermore, we note that at this stage, a comparison of our preliminary results with clinical gold standards is challenging. In fact, microvascular blood flow in the thyroid and nodules has not been thoroughly evaluated apart from only a pilot study utilizing arterial spin labelled magnetic resonance imaging, reporting significant changes in nodules [75,76].

Table 1. Variability of $BFi$ measured in thyroid.

View Table | View all tables in this article

Table 2. Variability of TRS measured parameters at all wavelengths in a single measurement in thyroid.

View Table | View all tables in this article

Table 3. Variability of TRS measured parameters at all wavelengths in a single day in thyroid.

View Table | View all tables in this article

Table 4. Variability of TRS measured parameters at all wavelengths between days in thyroid.

View Table | View all tables in this article

In the tables we have reported the coefficient of variation $CV$. The DCS repeatability tests highlight a variability of the single DCS acquisition of $3\%$ approximately, an inter-repetition variability of $10\%$, approximately the same as the variability between the $BFi$ values measured in different days. The variability of less than $3\%$ for a single DCS acquisition is considerably good in light of the low number of photons detectable from the thyroid tissue, characterized by high absorption [10,47]. This high quality result is made possible by the configuration of the probe - considering two identical source outputs placed in a symmetrical position with respect to detector fibers ($x - y$ in Fig. 1(b)), helping in increasing the photon count-rate - and by the possibility of averaging signals simultaneously acquired from 16 independent detection channels [53,54].

The TRS repeatability tests assess that the absorption coefficient has a single measurement variability $<2\%$, a variability between different measurements (probe replacement) $<4\%$, and a variability between different days $<10\%$. The reduced scattering coefficient shows a single measurement variability $<2\%$, a variability between different measurements (probe replacement) $<5\%$, and a variability between different days $<15\%$. In Fig. 4(d) the results of the repeatability tests for a single exemplary wavelength ($730\;nm$), for absorption (top panel) and reduced scattering coefficient (bottom panel) are shown.

3.5.2 Arm cuff demonstration

The capability of LUCA device in detecting changes in hemodynamics has been proved by measuring the arm muscle during an arterial cuff occlusion challenge. Since the LUCA device has been designed to perform measurements in stationary physiological condition, and does not have a high acquisition rate, we have decided to acquire with TRS just two wavelengths, $670$ and $830\;nm$, of the eight available, in order to be able to recover the oxy- and deoxy-hemoglobin concentration changes while maintaining an acceptable temporal resolution (roughly an acquisition every $7$ seconds). The LUCA probe has been placed in contact with the arm muscle, and a baseline signal has been acquired for roughly three minutes. Then we have applied a cuff pressure of $180\;mmHg$ for three minutes, and, lastly, we measured the recovery signal, releasing the cuff pressure, for roughly 5 minutes. In Fig. 5 we report the results of the arm cuff demonstration test. In panel (a) the relative blood flow index is reported, while in the (b) panel the oxy and deoxy hemoglobin changes are reported, for both the long and short source-detector distances. The high quality results reported demonstrate that the LUCA device is able to follow temporal hemodynamic changes with a high dynamic range, well in agreement with previous literature [35,61,70].

 

Fig. 5. Arm cuff occlusion demonstration test. Blood flow index ($BFi$) changes (left block) and oxy- and deoxy-hemoglobin change (right block)

Download Full Size | PDF

4. Discussion and conclusion

In conclusion, we have presented and characterized, through phantom and in vivo performance tests, the LUCA device.

The LUCA device is a unique optical-ultrasound clinical platform, developed in collaboration between academia, industry and clinicians, that combines hemodynamic and tissue constituent information, provided by TRS and DCS, in a way that could be incorporated within the standard, clinical thyroid cancer screening workflow, that is ultrasound imaging.

Overall, the LUCA device validation tests have demonstrated the capability of the system of performing high quality clinical measurements.

The optical subsystems are composed mainly by custom developed components that contribute significantly to reduce the costs with respect to standard commercially available ones, while maintaining very high performances. The DCS module, despite the reduction of costs due to completely new hardware components, is capable of performing high quality measurements even with low photon count-rates, allowing the simultaneous acquisition of $16$ independent detection channels. The TRS module consists of custom made laser heads and electronics, detectors and acquisition electronics, reducing the costs by a factor of 4-8 compared to equivalent commercially available components, while maintaining excellent performances.

The LUCA device’s clinical ultrasound imaging modality, in addition to providing anatomical information for the standard clinical thyroid cancer screening process, plays a role of guiding the probe on the right region of interest, reducing the in-determination due to probe misplacements for the optical measurements. In future, the anatomical US information can be used to further enhance the TRS/DCS analysis. Indeed, the LUCA device allows the radiologist to select the measurements of interest, draw different structures (e.g., trachea, thyroid, nodule, carotid) on the ultrasound images and then receive a report of analyzed data which is mimicked by this post-processing method. This is expected to improve the precision and accuracy of the optical data analysis and provide clinically useful measurement to enhance disease identification and prognosis. The use and effectiveness of these advanced methods requires detailed validation, and the correspondent study is still underway.

Concluding, the LUCA device was developed and tested for the clinical application in the field of endocrinology, where there is the necessity of improving the current standard of thyroid cancer diagnosis. Nevertheless, given the recent advances of diffuse optical techniques, the LUCA device can find potential application in several different fields, as for example breast, head and neck, abdominal cancers diagnosis and therapy monitoring, where the multi-modal approach involving simultaneous ultrasound imaging can produce a fundamental improvement in the quality of the examination.