Open Access
Add to BibSonomy
Abstract
For centuries, surgeons have relied on their sense of touch to identify vital structures such as blood vessels in traditional open surgery. Over the past two decades, surgeons have shifted to minimally invasive surgical (MIS) approaches, including laparoscopic surgery, which include benefits such as less scarring, less risk for infection, and quicker recovery times. In fact, some surgeries such as cholecystectomies have seen more than an 80% adoption of this technique because of those benefits. However, due to the fundamental challenges associated with using laparoscopic surgery, there has been a lower adoption in more complex specialties, such as colorectal and thoracic surgery, where the field of surgery has bleeding, fat, scar tissue, and adhesions. These problems are exacerbated by complicating factors such as inflammation, cancer, chronic disease, obesity, and re-operations. Importantly, surgeons will often convert from laparoscopy to open surgery if they can no longer proceed using the minimally invasive approach because of issues described with these complicating factors, thereby negating the benefits that the patient would have seen. When the surgeon does attempt these procedures with those issues, the surgery takes on average 30 min – 1 hour longer. A new method by which surgeons can visualize structures like blood vessels could reduce the conversion rates and operating time, thereby driving a greater adoption of laparoscopic surgery in these complex procedures. Here, we show that by adding near infrared (NIR) LEDs and a linear image sensor onto the opposing jaws of the laparoscopic graspers, blood vessels that are embedded within tissues can be detected and localized efficiently, even those not visible using current imaging techniques. We show the results of Monte Carlo simulations to support our claim, including that blood vessels ranging from 2 to 6 mm and buried under up to 1 cm of tissue can be detected. We also report developing a smart grasper handheld prototype to run ex vivo experiments. The results of these experiments matched with those of the Monte Carlo simulations and the estimated blood vessel size showed a strong correlation with the actual size. This technology will be incorporated into already existing laparoscopic tools to assist surgeons during MIS procedures.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Corrections
5 April 2018: A typographical correction was made to the abstract.
1. Introduction
With the increased use of MIS techniques during surgery, challenges in blood vessel detection and perioperative management of bleeding complications during these procedures have emerged [1,2]. The inability to palpate for the characteristic flow and pulsations within vessels makes identification highly dependent on surgeon experience [3]. Moreover, irregular vasculature that may exist due to congenital anomalies, previous operations, body habitus (e.g., obesity), or tumors, can complicate intraoperative dissection resulting in vascular injury [4–8]. Inadvertent bleeding causes non-reimbursable hospital costs amounting to billions of dollars in the US every year [9]. In the event of a significant bleeding complication during MIS, 32% of the cases result in death. The patients who survive these complications face an increased risk of developing an infection and a decreased quality of life [10]. To further increase the benefits of laparoscopic and robotic surgeries, which offer better surgical outcomes, quicker recovery times, decreased post-operative pain and reduced scarring at the surgical site, it will be essential to overcome the limitations in visualization posed by current MIS imaging capability. To prevent major complications, advanced tools must be created to supplement the techniques of surgeons when identifying the presence of blood vessels that must be avoided or strategically divided. While certain surgical specialties (i.e. neurosurgery and orthopedic surgery) have come to depend on stereotactic intraoperative localization, the vast majority of surgical procedures are undertaken without intraoperative guidance due to the non-stationary nature of the tissues involved [11–13]. Furthermore, all available modalities for intraoperative vessel visualization (i.e. ultrasound, fluorescence angiography) are insufficient based on one or a combination of the following: inherent technical limitations (i.e. resolution, speed, etc.) of the technology, inability to easily integrate into the surgical workflow, cost constraints, and/or added personnel requirements for equipment operation and/or analysis of test results [14–16]. Specifically, any technology that aims to enhance visualization of target tissues during surgery by taking advantage of the laparoscope must choose between a magnified, high resolution image or maintaining peripheral vision of the surgical field area. Available vasculature imaging technologies include: (1) laparoscopic ultrasound and (2) laser Doppler probes, which both lack sensitivity and are limited to detection of larger vessels. (3) Near-infrared (NIR) imaging has been pursued in breast reconstruction and colon surgery to offer information about the microvasculature, but offers low sensitivity and requires expensive contrast agents. The main limitations of existing vascular detection systems include the inability to quantitate vessel metrics, added complexity during surgery, and cost. We have developed a low-cost, real-time, and unobtrusive technology for vessel detection and quantification to address this need. The principle behind this technology is the optical attenuation of the light caused by the tissue and blood. The real world application of this principle has been used and validated by pulse oximeters, which are extremely common in the clinical setting. Pulse oximeters use the periodic absorption of red and NIR light, a result of the periodic change in the hemoglobin content, to predict the oxygen content of blood. In this paper we take the concept of pulse oximetry a step further and use it to detect and locate a blood vessel between the jaws, even through a thick layer of tissue. We also present the results of the Monte Carlo Simulations to support our claim. With the absence of human-induced artifacts, the results of these simulations are dependent upon the characteristics of the biological tissue and blood vessel of interest, giving us an absolute understanding of photon scattering and absorption inside the given tissue-vessel model. We also present the results from ex vivo studies to validate the real world application of this technology. The results from the Monte Carlo simulations and the ex vivo studies indicate that native blood vessels ranging in diameter from 2 to 6 mm can be visualized in up to 1 cm of tissue with a resolution of 1 mm.
2. Theory and methods
2.1. Pulse oximetry
Pulse oximetry is a technique to non-invasively measure oxygen saturation. It works by estimating the level of oxygenated and deoxygenated hemoglobin by light absorption. 810nm is an isosbestic wavelength, because light of that wavelength is absorbed to the same extent by oxygenated and deoxygenated hemoglobin. At either side of this isosbestic point, the absorption spectra of oxygenated and deoxygenated hemoglobin species invert, i.e. as the absorption of one species goes up, absorption goes down for the other species (shown in Fig. 1). Pulse oximeters utilize two LEDs, one red (660nm) and one NIR (910nm) wavelength, which are both passed through one side of a patient’s finger or ear and received by the photodetector on the other side. Scattering and absorption of light by the skin, tissues, and blood vessels (i.e. arterioles, venules and capillaries) attenuate the signal from the LEDs to the detector. The change in signal over time is recorded and quantified to estimate the oxygen content in that particular tissue.
Fig. 1 The absorbance spectra of oxygenated and deoxygenated hemoglobin. 810nm is known as the isosbestic point where oxy- and deoxy-hemoglobin exhibit the same absorption. These spectra are the foundation of pulse oximetry.
Pulse oximetry has been used in clinical settings for several decades and has proven the applicability of optics in detecting pulsatile flow of red blood cells. In this body of work, this concept is further developed to use absorption characteristics of tissues and vessels in order to detect, localize and characterize blood vessels during laparoscopic minimally invasive surgeries.
2.2. Instrument - smart handheld
To mimic the jaws of handheld surgical tools, a prototype with parallel jaws was created, whereby each jaw contained the required optical components: one jaw was equipped with an array of 5 NIR (λ = 940nm) LEDs (Epitex, Santa Clara, CA) each with a 34° solid angle. The other jaw contained a CMOS sensor array (Hamamatsu, Skokie, IL) embedded onto it (shown in Fig. 2, left). The sensor array had 2500 pixels, each with dimensions 0.125 × 0.007 mm2. To limit the computational processing requirements, the readings of 10 adjacent pixels were averaged to reduce the number of pixels from 2500 to 250. The equation below describes the averaging process where s denotes the readings of the 250 pixels that were created by averaging the 10 pixels (x) of the sensor array.
The spacing between any two consecutive LEDs was 3.2mm. The distance between the jaws, and consequently the LED and sensor arrays, can be modified using the mechanical trigger (shown in Fig. 2, right). Importantly, the power profile of the LEDs, currently being used at a 2.5% duty cycle, does not pose a burn risk to the tissue being studied. Even under direct, continuous contact with the tissue, the temperature of the tissue does not exceed the ambient temperature. In our ex vivo experiments, this temperature never exceeded 22°C with the LEDs shining at the brightest level and on for continuous 10 min. The output from the sensor array is relayed to a custom printed circuit board (PCB) containing a PIC32MZ microcontroller (Microchip, Chandler, AZ). The output from the CMOS sensor array is in the form of a continuous voltage signal ranging between [0, 3.3]V. This signal is sampled and digitized by a 12-bit Analog-to-Digital Converter (ADC) (S7476, Texas Instruments) added to the PCB, which governs the dynamic range of the system ([0, 212 − 1] or [0, 4095]). Herein, the dynamic range is defined as the range of possible output intensities based on a given LED brightness. The digital signal is then transferred to a custom MATLAB Graphical User Interface (GUI).
Fig. 2 The handheld prototype has a parallel jaw configuration, with an LED array and a sensor array on opposing jaws. The mechanical trigger is used to open or close the jaws, thereby adjusting the offset between the light source and detector. This prototype was designed for ex vivo feasibility testing of the vessel detection and localization algorithms.
The GUI is also used to drive the LEDs to the user-controlled brightness. The brightness of the LEDs was updated, depending on the tissue type and the size of the blood vessel between the jaws, in a way that the light intensity received by the detector stayed in a predefined range (See section 2.3.1 for details). The handheld prototype is used to assess tissues with embedded blood vessels.
2.3. Working principle
The readings from the linear image sensors on the lower jaw of the prototype create a spatial absorption profile (DC). With an inherent contrast between the absorption of the blood and that of the tissue, this DC profile could be used to detect blood vessels with (native) or without (skeletonized) the presence of any tissue around them (see Fig. 3, left). However, in the cases where the ratio of tissue thickness to the blood vessel size is higher, the DC alone might not be enough to differentiate a blood vessel from the surrounding tissue. For such scenarios there has to be an additional reference which could be used. As it turns out the blood vessels apart from being highly absorptive to NIR wavelengths have another natural contrast agent, the periodic blood flow. The light going through the blood vessel undergoes a periodic absorption over time because of such blood flow. The resultant periodic signal received by the sensor is commonly known as the photoplethysmogram (PPG). Pulse Oximetry is limited to a single sensor which captures time-only recordings; while our setup includes an array of sensors which allows spatial as well as temporal readings. The light intensity reaching the sensors beneath the blood vessel will be attenuated sinusoidally; and as a result the readings of those sensors will be periodic in time (shown in Fig. 3, right, red line). The sensors under the tissue alone will not see any such periodic change in the signal values (see Fig. 3, right, blue & green lines) due to absence of a blood vessel. This natural contrast in the light absorption properties of the flowing blood and the static tissue can then be used in conjunction with the spatial absorption profile along the sensor for a robust detection of the blood vessel. To quantify a change in the signal values over time, standard deviation is a commonly used statistical metric. The standard deviation, when computed for all the sensors in the array, gives another spatial profile which we call the AC-RMS profile. AC-RMS profile shows the region(s) of the sensor array which experience the most variation in the received light intensity. As mentioned earlier, the sensors underneath the blood vessel will record the most change. The AC-RMS profile for all the sensors except the ones underneath the blood vessel will be zero. A peak detection algorithm could be used to locate these sensors. In certain cases the external artifacts like motion can distort this AC-RMS resulting in an incorrect localization. These artifacts, however, do not affect the DC-profile as they do not create a significant drop in the light intensity so combining the AC-RMS profile with the DC profile can help in eliminating the effect of these artifacts. Importantly, while the water content of the tissue will lead to minor light absorption at 940nm, the light absorption by the blood in the vessels will be significantly greater due to the high concentration of hemoglobin, which has an absorption coefficient that is greater than that of water at 940nm by more than an order of magnitude [18]. In essence, while the AC-RMS and DC profiles have their limitations when used individually, their combination acts as a robust marker to detect, localize and quantify the size of the blood vessel.
Fig. 3 Left: DC profile. The sensors underneath the blood vessel experience a dramatic reduction in the light intensity caused by the high absorption characteristics of blood. The Right: AC-RMS profile. The absorption capacity of a given tissue stays constant over time, but the pulsatile nature of blood flow within a vessel causes the sensors beneath the blood vessel experience a sinusoidal signal over time.
Another factor that can lead to a periodic change in light absorbance is vessel compliance. The smooth muscle layer within the tunica media causes the vessel wall to dilate and contract each time the blood volume flowing through it increases or decreases. As the diameter of the blood vessel increases, it will cover more sensors, and as it decreases the number of sensors covered would decrease. As a result, the periodic change caused by the arterial compliance is spatial as well as temporal, making it an additional component of the AC-RMS profile.
2.3.1. Intensity adapt
The LED intensity i.e., the brightness of the LEDs is a critical element in understanding the periodic variations in the hemoglobin content of the blood. If the LED intensity is too low, the output signal will be below the noise level and will not contain any meaningful information. On the other hand, if this intensity is too high, the output signal will saturate the pixels deteriorating the resolution. It is therefore extremely important for the output intensity of the LEDs to be in a certain optimal range. Using the DC profile as a real-time feedback on the brightness of the LEDs, we can determine if the intensity of the LEDs at a given instant is sufficient to initiate the measurement and size estimation processes. To quantify this feedback, we are using the average of the DC profile and comparing that against the dynamic range of the system (See section 2.2). If this average is small (< 40% of the dynamic range) or large (> 75% of the dynamic range), the intensity of the LEDs is deemed to be low or high, respectively. The LED intensity is then increased or decreased to bring the average in the optimal range i.e., 40% < DC profile Average < 75%. The flowchart of this process is shown in Fig. 4.
Fig. 4 The flowchart describing the LED intensity adapt algorithm. Keeping the intensity in an optimal range is a critical factor for the size estimation methods to work well.
2.4. Monte Carlo simulations
For studying photon traversal inside biological tissues in a probabilistic manner, Monte Carlo simulations are generally considered to be a standard. A 3-D model of dimensions 19 × 19 × 10 mm3 was constructed, whereby a blood vessel of diameter xmm (x ∈ [2, 6]) was embedded within a tissue block of height 10mm. An LED array consisting of 5 LEDs with an inter-element spacing of 3.2mm was placed in the center of the top surface of the tissue block. As could be seen in Fig. 5, the sensor array was modeled to have 2500 pixels with dimensions matching the ones on the jaw of the handheld prototype and was placed in center of the bottom surface of the block. The Monte Carlo code was written in C and was based on the theory described in [17].
Fig. 5 A 3-D tissue block of size 19 × 19 × 10mm3 was modeled. An array of 5 LEDs with inter-element spacing of 3.2mm was at center of the top surface of the block (A). The light from all the LEDs undergoes some attenuation that is dependent on the optical characteristics of the tissue(s) and size of the blood vessel traversed by the photons. The attenuated light that reaches the other side is recorded by the sensors in the linear array under the tissue block (B). The block is inverted upside down (B) to show the placement of the sensor array under the tissue block.
The irradiated light from the LEDs was simulated by launching a few billion photons inside a block of tissue with and without a blood vessel embedded in it. The photons were launched in packets, and each packet was tracked individually as it moved through the block. With each movement of the photon packet, its direction and weight was updated depending on the scattering and absorption properties of the medium. The photon was tracked until its weight was above a certain threshold (1% of its initial weight) and/or it escaped from any surface of the tissue block. A flowchart of the entire process is shown in Fig. 6. The coordinates of the sensors in the sensor array were computed using basic geometric and trigonometric identities. Once the coordinates of the sensors were known, an array of equal size to the number of sensors (N) in the array was initialized with a weight of zero. The weights of the sensors denote the amount of current generated by them. When no photons have yet reached a sensors within the array, the current for that sensor is non-existent and therefore the sensor has a weight of zero. Block k in the array corresponded to sensor k in the array, with known coordinates. The coordinates for the endpoints of the entire sensor array were computed using the size of a single sensor and the number of sensors in the array. Only the photons that reached the coordinate region of the sensor array were used to populate the signal. The exact coordinate helped in determining the sensor number that absorbed the incoming photon. Once the simulation was complete, the photons absorbed by each sensor were counted and that quantity was used to construct the spatial absorption profile or the DC profile across the sensor array. To validate the ability of the AC-RMS signal to be used as a potential marker for blood vessels, we simulated cases with different blood vessel diameters and with and without surrounding tissue. The blood flow through the vessel and diameter were varied by changing hemoglobin content and diameter in a sinusoidal fashion. A single simulation of the Monte Carlo is generally not known to provide the temporal information. In this paper, each simulation of the Monte Carlo provides a snapshot of a given time instant. That is, to simulate the temporal changes, we ran multiple (n = 10) simulations where the values of parameters like μa and the blood vessel diameter were updated in a sinusoidal manner while keeping the number of launched photons the same. As an example, the diameter of the 3mm blood vessel was varied across the 10 simulations as [3mm, 2.52mm, 2.26mm, 2.35mm, 2.74mm, 3.25mm, 3.65mm, 3.74mm, 3.48mm, 3mm]. This was done to mimic the arterial compliance of a vessel during blood flow. As the blood pulses through a blood vessel, the diameter and the overall absorption of the light follows a periodic curve. The results of these 10 simulations were put together in a matrix of size 10 × 250. The columns of the matrix represented the 250 sensors and the rows contained the output of these sensors at each time instant (each simulation). The AC-RMS profile is then computed by taking the standard deviation of these 10 readings at each sensor. The AC-RMS and the DC profiles, when plotted together, show a common pattern that could be used to detect and localize a blood vessel (see Fig. 8 – 9).
Fig. 7 Left: The spatial absorption profiles (DC) of blood vessels ranging in diameter from 2mm to 6mm. There is a clear relation between the size of the blood vessel and the widths/depths of DC profiles. The width of these dips can be used as a marker for the size of the corresponding blood vessels. Right: The spatial absorption profiles (DC) of blood vessels ranging in diameter from 2mm to 6mm embedded in an adipose tissue block. The light attenuation caused by tissue is clearly visible as the DC profile loses its flatness on the sides.
Fig. 8 The AC-RMS (red) and DC (blue) profiles together can be used as a marker to detect and localize skeletonized blood vessels. The peaks in the AC-RMS profile denote the region of the arterial compliance. The modeled compliance is higher in the 6mm blood vessel (right) than in the 3mm vessel (left), resulting in bigger AC-RMS peaks for the larger blood vessel.
Fig. 9 The AC-RMS (red) and DC (blue) profiles together can be used as a marker to detect and localize native blood vessels. The peaks in the AC-RMS profile denote the region of the arterial compliance. The modeled compliance decreases due to the tissue, however retains a higher magnitude in the 6mm blood vessel (right) than in the 3mm vessel (left), resulting in bigger AC-RMS peaks for the larger blood vessel.
2.5. Ex vivo setup
Carotid arteries (n = 12, ranging in diameter from 3mm–7.2mm) were excised from fresh porcine tissue specimens the day of each experiment, and threaded onto tube fittings (McMaster-Carr, Elmhurst, IL). These arteries were then fitted to 1/8″ PVC tubing (McMaster-Carr, Elmhurst, IL) at each end, with the proximal end connected to a custom pump. A custom pulsatile system was employed to mimic cardiac-induced blood flow through ex vivo porcine arteries in a series of bench-top experiments. A VPII submersible DC pump (Simply Pumps, Manor, PA) controlled by a Teensy 3.1 microcontroller board was used to simulate pulsatile arterial flow with physiological blood volume equivalent to the hepatic artery, and heparinized porcine blood (98% oxygenation) was pulsed through the pump to in-line ex vivo porcine carotid arteries. The pump rate was kept at 100 cycles per minute, to match the average resting porcine heart rate. Blood temperature was maintained at physiological temperatures (37°C) to ensure proper viscosity. For all experiments, the artery was placed directly onto the sensor array (bottom jaw of the device), and the top jaw of the handheld was kept at a constant distance from the sample (1mm from the highest point of the artery sample). The grasper force was the same across all the experiments, as the grasper was kept in a fixed position for experiments. The intensity adapt method (optimal intensity selection method as described in section 2.3.1) was used to keep the level of the DC profile the same across different experiments. The LED intensity adapt process compensated for the loss of light caused by specular reflection and other attenuations. The dynamic range of the handheld was [0, 4095] as it had a 12-bit ADC. A custom MATLAB GUI (Mathworks, Natick, MA) was used to capture the signal and to analyze the DC signal and signal variance profiles created by the pulsing artery. The vessel diameter estimates captured using the system were plotted against gross diameter measurements of the corresponding, blood-filled arteries taken using a digital caliper (Whitworth). The sample data was used to compute the linear correlation coefficient (R2), and a T-Test was performed to calculate the significance of the regression coefficient. Although the variance seen in the ideal system of the Monte Carlo simulations was extremely small (on the order of 10−7), we assumed a variance of 0.5 in the real world ex vivo samples, and used this variance in a power analysis (α = 0.05, power= 0.90, σ = 0.5) to calculate the sufficient sample size required to find a statistically significant 1-mm difference in vessel diameter for our experiments. This power analysis determined that n = 6 was sufficient for significant results, and we did achieve significant results within our first 6 samples. However, as we had access to the porcine tissues as a waste product from a slaughterhouse, we saw it fit to continue to collect data beyond n = 6.
3. Results and discussion
The Monte Carlo simulations were performed for different blood vessels with outer diameter ranging from 2mm to 6mm, with and without surrounding tissue. For each experiment, we launched 2 billion photons inside the tissue block. Adipose tissue (μs = 122.9cm−1, μa = 0.13cm−1 and g = 0.9, n = 1.4 at λ = 940nm [18]) was used for the initial analyses, with a light source consisting of an array of 5 940 nm LEDs with a solid angle of 34°. The detector used was a linear sensor array with 2500 pixels, each of which had dimensions of 0.125 × 0.007mm2. For the simulations with the blood vessel, it was kept resting on the sensors. The optical properties for the blood used were μs = 53.2cm−1, μa = 6.62cm−1, g = 0.9, n = 1.3 at λ = 940nm [18]. The DC profiles obtained from these simulations were divided into two parts: regions with blood vessel and regions without a blood vessel. The averages of the DC readings of the latter regions were used to normalize the entire DC profile. To match the dimension of these DC profiles with the ex vivo settings, they were further averaged over 10 consecutive sensors to reduce the number of pixels from 2500 to 250 (as described in section 2.2).
3.1. Spatial absorption (DC) profiles
The initial Monte Carlo simulations included a skeletonized blood vessel model to determine the relationship between DC profiles and sizes of the corresponding blood vessels. DC profiles were computed for blood vessels with varying sizes between 2 and 6 mm. (Fig. 7, left). The heavy and symmetric dip in the middle was a result of the heavy absorption of light by the blood and indicates the presence of the blood vessel in the center of the sensor array. The simulations demonstrate that with increasing diameter of blood vessels, the resulting DC profile is deeper and wider. As a result, not only these symmetric dips could be used to detect the presence of a blood vessel but the width of these dips can be used as a marker of its size.
To see if the same pattern can be obtained when the blood vessel is surrounded by tissue, the simulations were repeated with the same blood vessels embedded within 1cm of adipose tissue. Once the blood vessel is inside a tissue block, the flatness around the DC profile is lost (Fig. 7, right). Attenuation of light by the tissues is apparent in its effect on the spatial absorption profile.
With increased variability around the blood vessel, i.e., heterogeneous tissue thicknesses, perfusion, etc., the spatial absorption profiles across the sensor array are vulnerable to distortion. Under these circumstances, the width of the dip might not remain a good marker for the size and for that reason we have combined this marker with the AC-RMS that would serve as an universal marker for native and skeletonized tissue.
3.2. Integrating temporal information (AC-RMS) with spatial profile (DC)
To validate the ability of the AC-RMS signal to be used as a potential marker for blood vessels, we simulated cases with different blood vessel diameters and with and without surrounding tissue.
The blood flow through the vessel and diameter were varied by changing hemoglobin content and diameter in a sinusoidal fashion. The diameter was changed to mimic the arterial compliance during blood flow. The AC-RMS and the DC profiles, when plotted together, show a common pattern that could be used to detect and localize a blood vessel (Fig. 8).
The next set of simulations were performed in the presence of surrounding adipose tissue. The blood vessels were kept at 1cm below the surface of adipose tissue, and the simulation was performed as described above. The tissue affects the compliance motion of the arteries, as it does not allow the artery walls to extend as much as they do in the absence of tissue. As a result, the magnitude of the peaks in the AC-RMS decreases (Fig. 9). However, the relative positioning of the peaks remains about the same, indicating that the arteries can still be detected. The distance between the two peaks also acts as a marker for the corresponding blood vessel size. The Monte Carlo simulations demonstrated that by combining the DC and AC-RMS profiles, it is possible to detect and quantify a blood vessel even if it is embedded within tissue. Even though the DC profile degrades based upon the tissue thickness surrounding the blood vessel, the AC-RMS stays relatively in the same place. While Monte Carlo simulations support the combined use of DC and AC-RMS, we have further validated this combined marker in ex vivo settings using porcine blood vessels.
3.3. Ex vivo results
The artery was placed between the jaws of the handheld prototype (Fig. 2), and the custom GUI was used to capture the signal created by the transmitted light at the CMOS detector. The DC and AC-RMS profiles created by the pulsing artery were analyzed to estimate vessel size for comparison to measured diameter values of each artery. The DC and AC-RMS profiles obtained in the experiment setup match the profiles obtained during the Monte Carlo simulations (Fig. 10, left). The DC profile was normalized using the method described in section 3. It can be seen that because of the periodic absorption of the light by the blood flow, the sensor below the blood vessel gives out a sinusoidal signal (PPG) (Fig. 10, right). As was observed in the Monte Carlo simulations, the other sensors that are not under a blood vessel do not experience any pulsatile change in the absorption and record an almost constant signal over time. Using the DC and the AC-RMS profiles we were able to detect the presence of a blood vessel and localize its position along the sensor array. The distance between the peaks of the AC-RMS (shown as AC width in Fig. 8 and Fig. 9) was used to estimate its size. The size estimates computed by the system correlated well (Fig. 11) with the gross diameter measurements, taken with a digital caliper, for each of the arteries (R2 = 0.96, p < 0.02), with an average error of 0.19mm. This confirms the ability of the system to work with high fidelity to detect and quantify the size of native blood vessels under ex vivo conditions.
Fig. 10 (Left) The DC (blue) and AC-RMS (red) profiles from ex vivo experiments. The profiles match the ones acquired in the Monte Carlo simulations. (Right) The signal from the sensors next to the blood vessel have an almost constant signal (blue and red lines). One of the sensors below the blood vessel records a PPG signal (yellow).
Fig. 11 The results obtained from ex vivo studies. The size estimated using a combination of the DC and the AC-RMS profiles showed a strong correlation between the actual and the estimated sizes of the blood vessels. Our method was able to quantify the skeletonized and native blood vessels with an average error of 0.19mm in the ex vivo settings.
4. Conclusion
We have demonstrated that skeletonized and native blood vessels can be detected and localized using a handheld prototype with LED and sensor arrays on the opposing jaws. Monte Carlo simulations indicate that a blood vessel even buried under 1cm of tissue could be detected efficiently. We introduce the spatial DC and the AC-RMS profiles which are different for blood vessels with different diameters and show that a combination of the two profiles could be used to estimate the size of the blood vessels. In addition to MC simulations, we conducted ex vivo studies, wherein the system was able to estimate the sizes of blood vessels with a high accuracy for blood vessels ranging from 3 – 7mm in size. Importantly, the system was able to clearly visualize the location and give accurate diameter estimates of native blood vessels, which is not possible using existing surgical graspers, or any other intraoperative imaging modality currently available. In future studies, we will be testing the handheld prototype in vivo to validate the concept. We will also be working towards miniaturizing this handheld to mimic the laparoscopic graspers available on the market. The performance of the updated prototype will be compared with the other existing cumbersome techniques.
Funding
National Science Foundation Small Business Innovation Research Program (520502 and 1660240).
Acknowledgments
The results presented here are based upon work supported by the National Science Foundation Small Business Innovation Research program grants 1520502 and 1660240.
The authors would like to acknowledge that the lab support was provided by the Insight Product Development, Chicago.
Disclosures
AC: Briteseed (I,E,P), SAS: Briteseed (I,E,P), PLR: Briteseed (I,E,P), MV: Briteseed (I,E,P), JWG: Briteseed (I,E,P), HS: Briteseed (I,C,P)
References and links
1. J. F. Magrina, R. Cetta, Y. Chang, G. Guevara, and P. M. Magtibay, “Analysis of secondary cytoreduction for recurrent ovarian cancer by robotics, laparoscopy and laparotomy,” Gynecologic Oncology 129, 336–340 (2013). [CrossRef] [PubMed]
2. E. David-Montefiori, R. Rouzier, C. Chapron, and E. Darai, “Surgical routes and complications of hysterectomy for benign disorders: a prospective observational study in French university hospitals,” Human Reproduction 22(1), 260–265 (2007). [CrossRef]
3. N. J. Crane, B. McHone, J. Hawksworth, J. P. Pearl, J. Denobile, D. Tadaki, P. A. Pinto, I. W. Levin, and E. A. Elster, “Enhanced surgical imaging: laparoscopic vessel identification and assessment of tissue oxygenation,” J. Am. Coll. Surg. 206(3), 1159–1166 (2008). [CrossRef] [PubMed]
4. R. L. Prosst, E. D. Fernandez, W. Neff, C. Braun, T. Neufang, and S. Post, “Evaluation of MR-angiography for preoperative assessment of living kidney donors,” Clin. Transplant 19, 522–526 (2005). [CrossRef] [PubMed]
5. A. Rafii, S. Camatte, L. Lelievre, E. Darai, and F. Lecuru, “Previous abdominal surgery and closed entry for gynaecological laparoscopy: a prospective study,” BJOG: an Int. J. of Obstetrics and Gynaecology 112, 100–102 (2005). [CrossRef]
6. K. Martay, G. Dembo, Y. Vater, K. Charpentier, A. Levy, R. Bakthavatsalam, and P. R. Freund, “Unexpected surgical difficulties leading to hemorrhage and gas embolus during laparoscopic donor nephrectomy: a case report,” Can. J. Anesth , 50, 891–894 (2003). [CrossRef] [PubMed]
7. I. S. Gill, L. R. Kavoussi, R. V. Clayman, R. Ehrlich, R. Evans, G. Fuchs, A. Gersham, J. C. Hulbert, E. M. McDougall, T. Rosenthal, W. W. Schuessler, and T. Shepard, “Complications of laparoscopic nephrectomy in 185 patients: a multi-institutional review,” J. Urol. 154, 479–483 (1995). [CrossRef] [PubMed]
8. G. Pareek, S. P. Hedican, J. R. Gee, R. C. Bruskewitz, and S. Y. Nakada, “Meta-analysis of the complications of laparoscopic renal surgery: Comparison of procedures and techniques,” J. Urol. 175, 1208–1213 (2006). [CrossRef] [PubMed]
9. Assuming a 1% incidence rate (see Opitz, et. al.) and $210K cost per incidence based upon review of Northwestern Memorial Surgical Outcome Data for 2011 analyzing surgeries coded with excessive hemorrhaging.
10. I. Opitz, W. Gantert, U. Giger, T. Kocher, and L. Krahenbuhl, “Bleeding remains a major complication during laparoscopic surgery: analysis of the SALTS database,” Langenbeck’s Archives of Surgery 390(2) 128–133 (2005). [CrossRef]
11. H. Iqbal and Q. Pan, “Image guided surgery in the management of head and neck cancer,” Oral Oncol , 57, 32–39 (2016). [CrossRef] [PubMed]
12. R. L. Satcher Jr., “How intraoperative navigation is changing musculoskeletal tumor surgery,” Orthop. Clin. North Am. 44(4), 645–656 (2013). [CrossRef] [PubMed]
13. S. Atallah, G. Nassif, and S. Larach, “Stereotactic navigation for TAMIS-TME: opening the gateway to frameless, image-guided abdominal and pelvic surgery,” Surg. Endosc. 29(1), 207–211 (2015). [CrossRef]
14. A. S. Sethi, S. M. Regan, and C. P. Sundaram, “The use of a Doppler ultrasound probe during vascular dissection in laparoscopic renal surgery,” J Endourology 23(9), 1377–1382 (2009). [CrossRef]
15. P. Newman and G. Rozycki, “The history of ultrasound,” Surgical Clinics of North America 78(2), 179–195 (1998). [CrossRef] [PubMed]
16. J. L. Saumet, D. L. Kellogg Jr, W. F. Taylor, and J. M. Johnson, “Cutaneous laser-Doppler flowmetry: influence of underlying muscle blood flow,” Journal of Applied Physiology 65.1, 478–481 (1988). [CrossRef] [PubMed]
17. L. Wang, S. L. Jacques, and L. Zheng, “MCML-Monte Carlo modeling of light transport in multi-layered tissues,” Computer Methods and Programs in Biomedicine 47(2), 131–146 (1995). [CrossRef] [PubMed]
18. S. L. Jacques, “Optical properties of biological tissues: a review,” Physics in Medicine & Biology 11(58), R37 (2013). [CrossRef]
