Ex Vivo Evaluation of a Soft Optical Blood Sensor for Colonoscopy

Summary

Colonoscopies are vital procedures allowing diagnosis of colorectal cancer and other gastrointestinal diseases. However, excessive forces may be applied to the colon during navigation. This can cause bleeding, especially in patients presenting inflammatory bowel diseases. The endoscopist is often unable to detect bleeding as visualization is limited to the distal tip camera of the endoscope. Thus, there is a need to have bleeding detection capabilities behind the device tip. This work presents a soft optical blood sensor that can be mounted onto a colonoscope. The presence of blood in the sensor’s microchannel causes a reduction in optical transmission, and the endoscopist is alerted. We evaluate the sensor safety and performance ex vivo with a cohort of 10 endoscopists (novices and experts). We demonstrate the ability of the sensor to rapidly identify bleeding and easily integrate into the clinical workflow, without significantly affecting navigation time and the users’ learning curve.

Graphical Abstract

eTOC Blurb:

This work presents a soft optical blood sensor that can enable real-time blood detection behind the colonoscope’s camera. The sensor was evaluated in a user study to assess the device’s ability to identify blood in a real colon, as well as understand its impact on the clinical workflow. The sensor was shown to integrate smoothly into the colonoscopy workflow without increasing the time taken for the procedure or affecting the total workload on the surgeon.

1. Introduction

Colorectal cancer is a leading cause of cancer-related death worldwide and accounts for more than 1.85 million cases annually¹. In the US alone, there will be an estimated 153,020 new cases of colorectal cancer and result in 52,550 deaths in 2023². Colonoscopies are vital procedures to detect and address incidences of colorectal cancer³. Additionally, colonoscopies also allow screening and detection of other severe gastrointestinal (GI) diseases such as inflammatory bowel diseases (IBD), including Crohn’s disease and ulcerative colitis, large bowel obstruction (LBO), diverticulitis, and angiodysplasia^4–7. IBD is estimated to affect 1.3 % of the entire US population⁸ while diverticulitis resulted in over 2.7 million hospital visits⁹. Thus it is critical that patients undergo regular colonoscopic procedures for timely screening and proper treatment of various GI conditions. Colonoscopy often requires the endoscopist to maneuver the colonoscope through tortuous regions, such as the sigmoid colon¹⁰. The flexibility of the colonoscope makes it prone to looping¹¹ and the presence of friction causes inaccurate transmission of forces from the contacted colon wall to the endoscopist’s hands¹². This could lead to potentially excessive application of forces^13,14. Excessive forces often result in severe adverse events (SAEs) that include perforation, bleeding, and splenic injury¹⁵. More than 15 million colonoscopies are performed annually, of which the most frequent SAE is bleeding, which is present in 6.5 to 23.1 cases per 10,000 procedures¹⁶. Additionally, the incidence of SAEs is quite variable, depending on the endoscopist, nature of the procedure, and patient characteristics¹⁷. The likelihood of colorectal bleeding may be exacerbated due to conditions such as polyps, hemorrhoids, or infected diverticula^7,18–21. Colorectal bleeding can often occur when the colonoscope becomes looped and presses onto the colon wall¹⁰. Since the endoscopist is only able to visualize the anatomical environment through the camera at the distal tip, any iatrogenic bleeding (i.e., bleeding induced by the procedure) behind the distal tip may not be detected in a timely manner. Additionally, colonoscopies present significant limitations in definitely identifying sources of non-iatrogenic lower gastrointestinal bleeding (LGIB) after it has occurred. As per one study, only 33 % of LGIB bleeding sources were accurately identified²².

To address the issue of bleeding, colonoscopes must be able to identify bleeding in real-time beyond just the distal tip. Prior work in bio-sensing and blood detection have largely focused on characterizing blood composition (i.e, cell or platelet counting) or bio-marker detection^23–26 as opposed to whole blood detection. These consist of lab on chip (LOC) devices that utilize microfluidic cytometry^27–32. However, the complexity and rigidity of such systems impedes integration onto flexible surgical tools, i.e., colonoscopes, that need to navigate through the GI tract and require rapid, real-time blood detection.

To specifically address real-time GI bleeding, there have been recent developments in wireless capsule endoscopy (WCE). The devices can be swallowed and passed through the GI tract through the digestion process while providing imaging of the tract. WCE devices utilize cameras and CMOS sensors to map the gastrointestinal tract and identify bleeding spots. However, most WCE devices focus on acute bleeding in the upper GI tract or small bowel as opposed to the colorectal region^33–35. Additionally, WCE devices are standalone sensors that cannot be integrated onto traditional endoscopes and are meant to diagnose pre-existing bleeding. In traditional interventional endoscopy, bleeding is often induced by interaction of the colonoscope itself with the colon. Thus, there is a need to augment blood sensors onto traditional colonoscopes as opposed to stand-alone, wireless capsule devices.

In our prior work, we developed a proof-of-concept soft optical sensor that can detect blood in real-time and validated it in an in-vitro mock colon³⁶. The sensor consists of a soft polymer cladding and soft optical waveguides that interface with a microchannel that can intake blood. The presence of blood is identified by optical absorption of the incident light. This can be measured as an optical loss by the opto-electronic equipment.

Although the sensor was able to detect simulated bleeding in a colon artificial model, the sensor’s signal was also susceptible to false positives due to deformations (caused by incident forces between the colonoscope and the colon and colonoscope bending). In order to translate the sensor closer to clinical use, the sensor required significant changes to its design and had to be assessed in a surgical environment utilizing real tissue models.

In this work, we advance the design and clinical functionality of the soft optical blood sensor (Figure 1), evaluating the device through an ex vivo user study involving multiple endoscopists of varying skill levels, ranging from novices to experts. The sensor’s physical performance is evaluated using metrics such as time taken to detect bleeding, the sensor’s ability to safely navigate through the colon without tissue perforation, and the sensor’s capacity to withstand fatigue from prolonged usage. We also present a human factors analysis of the sensor to ascertain its impact on user workload when integrated into the clinical workflow. This is performed via a NASA Task Load Index (TLX) workload study and learning curve analyses of navigation time.

1. Blood sensor overview.

A) Diagram of the issue of bleeding induced behind the colonoscope’s distal tip. The endoscopist cannot identify bleeding due to lack of visualization. B) An illustration depicting the soft blood sensor mounted onto a colonoscope. As the colonoscope navigates through the colon with bleeding present, the blood enters the sensor microchannel and attenuates incident light from the waveguides. This allows the sensor to detect blood and inform the endoscopist to investigate further. (C) CAD rendering of the sensor showing the microchannel and the optical waveguides that transmit green light and infrared light for selective identification of blood. D) Image of blood sensor mounted on a colonoscope, with the blood analog in the microchannel. Visible (green) and infrared (blue) waveguides are highlighted. E) Microscope image showing the interfacing of the optical waveguides with the microchannel at the optical interrogation zone. F) Microscope image displaying the transmission of light across the microchannel filled with an illuminating tracer dye. The micro-lens is shown focusing the incident visible light.

The design of the new soft optical sensor addresses limitations in our prior work³⁶ to function effectively in an unstructured surgical environment (i.e., an explanted bovine colon) close to a real clinical setting. The sensor can filter out signals from deformations, including contact forces and bending encountered in the colon during navigation. This is achieved with multiple pairs of optical waveguides and the use of both visible and infrared light (Figure 1C). Physical deformations such as incident force cause light in both waveguides to attenuate. This is due to the close proximity of the infrared and visible waveguides, ensuring that the resultant deformation of the sensor impacts both pairs of waveguides. However, the presence of blood attenuates green light to a greater degree than the infrared light, as a result of its hemoglobin content³⁷. This difference in attenuation between visible and infrared wavelengths can be measured as a net optical loss, allowing the sensor to differentiate between physical deformations and bleeding. In this work, the blood sensor is evaluated using a blood analog that mimics the optical characteristics of actual blood (see Video S1, Supplemental Information). Additionally, the sensor incorporates an optical adhesive as a waveguide core material. The waveguide adheres to a commercial glass optical fiber that transmits light from the opto-electronic circuit (Figure 1C inset). The chemical adhesion helps reduce sensor fatigue and detachment of the fibers during endoscopic navigation. The sensor is connected to a fluidic system, allowing the sensors to sequentially vacuum fluid in the colon to detect bleeding. The fluidic system also allows the sensor to flush any clogged mucus from the microchannels using saline solution. The sensor data is relayed through a graphical user interface that alerts the endoscopist to the specific sensor that detects the bleeding during the colonoscopy.

2. Results

2.1. Sensor Design

The soft optical sensor consists of a silicone cladding of Polydimethylsiloxane (PDMS) with four curved, parallel waveguides that interface with a central microfluidic channel (Figure 1C–D). The sensor contains a 4 mm diameter blood intake pool that connects to a rectangular fluidic channel measuring 100 μm in width and 200 μm in height (Figure 1D). The rectangular channel passes through the optical path of the waveguides. This forms a narrow, optical interrogation zone for the incident light from the waveguides. The interrogation zone is seen in (Figure 1E–F), where the incident light is illuminated with a tracer dye (FWT Red 25, Kingscote Chemicals). The channels then diverge to a width of 1 mm. This facilitates the attachment of tubing that connects to a vacuum pump, enabling suctioning of GI fluids and blood. If the channel clogs due to mucous from the colon sub-mucosal wall, it is cleared using a secondary in-line pump that flushes pressurized air or saline solution. The sensor also contains a plastic base (not shown in figure) and is mounted onto a flexible resin sleeve (not shown in figure). This acts as a strain limiting layer that protects the sensor from excessive bending and deformation when the colonoscope loops during navigation. A total of three sensors can be attached radially onto the sensor sleeve, which is then mounted onto the colonoscope. This allows bleeding detection at three locations around the circumference of the device. The sensor measures 10 mm in width by 25 mm in length, with a vertical thickness of 1 mm. The low vertical profile of the sensor allows integration onto existing colonoscopes without adding excessive radial thickness to the device, which may impair insertion and navigation within the colon. Additionally, the sensor is designed to serve as a single-use device. It is fabricated using low-cost, flexible materials that allow the sensor to be easily disposed of after the colonoscopy. This eliminates any issues concerning the sterilization of the device between procedures.

2.2. Waveguide Design and Sensing Principle

The optical waveguides consist of a polymeric core material with a high refractive index (NOA 73, Norland Inc.) encapsulated by a lower refractive index cladding (PDMS). The difference in refractive index between the core (n₁ = 1.56) and the cladding (n₂ = 1.43) allows incident light to travel through the waveguide by total internal reflection (TIR). Light rays can propagate through the waveguide if they have an angle of incidence above a critical angle of 66.4°. The critical angle θ_c of the waveguide-cladding interface is calculated as shown in Equation 1:

$${\theta }_{\text{c}}={\text{sin}}^{-1}\left({\text{n}}_2/{\text{n}}_1\right).$$ (1)

The outer waveguides transmit visible green light (525 nm) while the inner waveguides transmit infrared light (950 nm). Since whole blood has absorbance peaks between 500 nm to 600 nm due to the presence of hemoglobin, the visible green light is attenuated by the blood while the infrared light experiences minimal attenuation³⁷. This produces a unique spectral response which allows the sensor to identify blood from physical deformations, which attenuate both wavelengths of light. The absorbance of visible light by blood is interpreted as a drop in signal voltage by the opto-electronic circuit. Thus, the change in voltage with respect to the initial baseline voltage can be measured as an optical power loss using Equation 2:

$$\text{Loss}(\text{dB})=10{\text{log}}_{10}\left(\frac{{\text{I}}_0}{\text{I}}\right)$$ (2)

where I_o and I are the baseline and output power of the light across the waveguides and the loss is measured in decibels.

The input waveguide transmits visible light across the microfluidic channel into the receiving waveguide via an in-plane microfludic lens (Figure 1E). The the micro-lens confine divergent light rays that are emitted from the transmitter waveguide and couple it into the receiving waveguide across the micro-channel. This reduces coupling losses and improves the signal baseline, thereby optimizing the sensor’s performance (see Note S1 and Figure S1, Supplemental Information). The waveguides are curved concentrically in a 90° bend (Figure 1E–F) with a cross section of 200 μm. The radius of curvature (ROC) of the outer, visible light waveguide is 2 mm, while the inner waveguides have a ROC of 1 mm. This allows the visible and infrared waveguides to be situated in close proximity to each other. Thus, localized physical deformations, such as bending and incident force on the sensor, result in scattering and attenuation of light across both pairs of waveguides. This affects the both wavelengths of light and results in similar optical losses in the two pairs of waveguides. Consequently, the optical loss across each waveguide can be subtracted and normalized to filter out unwanted signals from physical deformations. This is seen in the equation below:

$$\text{Visible}\text{Loss}(\text{dB})-\text{Infrared}\text{Loss}(\text{dB})={\log }_{10}\left(\frac{{\text{I}\text{vis}}_0}{\text{I}\text{vis}}\right)-{\log }_{10}\left(\frac{{\text{I}\text{ir}}_0}{\text{I}\text{ir}}\right)$$ (3)

where I_o and I are the baseline power and output power of the visible and infrared light waveguides respectively. Therefore, the presence of blood causes the visible light to be attenuated which results in a net positive relative loss, while physical deformations will cause a net zero relative loss.

2.3. Blood Concentration Sensor Loss Characterization

The sensor’s optical response was evaluated with respect to variations in concentration of the blood analog (Carolina Biological Supply Company). The sensor must be able to detect a range of blood concentrations as blood in the colon may get diluted due to the presence of gastrointestinal fluid. Low concentrations of blood may not be detected due to reduced attenuation of incident light. The attenuation of incident visible light is modelled by the Beer-Lambert law as shown below:

$$\text{A}={\log }_{10}\left(\frac{\text{Io}}{\text{I}}\right)=\epsilon \text{lc}$$ (4)

where A represents the absorbance which can be represented as the logarithm of the baseline power (I_o) and output power (I) respectively, ε is the Molar absorption coefficient dependent on the attenuating substance, l is the optical path length, and c is the molar concentration of the attenuating substance. A higher blood concentration results in a higher molar concentration of the attenuating substance, as seen in Equation 3. Consequently, the absorbance of the solution increases. This reduction in light transmission causes a lower output power with respect to the baseline power. This is recorded by the photo-receiver as an optical power loss. Conversely, the blood does not absorb and attenuate the infrared light, and there is a slight optical “gain”. This is because the refractive index of the blood analog is higher than air. Thus, a lower proportion of the incident light is scattered by the medium due to lower Fresnel reflection losses. This causes an increase in the light transmission of the IR waveguide when the blood analog is present. The distinct difference in optical absorption of the blood analog to visible and infrared light can be observed in Figure 2A. In our prior work³⁶, the blood sensor utilized a single wavelength of 525 nm green light. The sensor was susceptible to loss from deformations, hence spikes in loss caused by physical could be falsely interpreted as bleeding. By utilizing both visible and infrared light, the presence of blood elicits a unique spectral response from the sensor compared to physical deformations. This allows the sensor to selectively distinguish bleeding from other fluids such as gastrointestinal fluid (see Note S2 and Figure S2, Supplemental Information) and mitigate false detections from deformation.

2. Optical loss behavior of the blood sensor in response to blood analog.

A) Sensor shows a dual-response to the presence of blood in the microchannel. Visible light is attenuated causing an optical loss while infrared light is unaffected. B) Visible light optical response of the blood sensor with respect to the blood analog ranging from 20% to 100% concentration (v/v). The solid line is the mean value and the shaded area represents the standard deviation computed on four prototypes.

The optical loss behaviour of the sensor’s visible light waveguides is illustrated in Figure 2B. As the concentration of blood increases, the visible light optical loss increases in a linearly manner. The sensor exhibits a range of sensitivity of 1.3 dB to 5.6 dB as the blood concentration increases from from 20% to 100%. The detection threshold of the optical loss in the control algorithm is set to 2 dB. This means that low concentrations of blood below 30% are selectively filtered out and will not alert the endoscopist of potential bleeding. This threshold was set at 2 dB to avoid the possibility of false detections from other phenomena. For example, minor bleeding, consisting of a low volume of highly diluted blood, may cause a slight rise in loss. Additionally, small bubbles may be present in the fluid flow through the microchannel which may cause brief, low magnitude spikes in loss (< 2 dB). If the bleeding is significant, the endoscopist will be immediately alerted as optical loss exceeds the set threshold (i.e., 2 dB). This is irrespective of the exact concentration and magnitude of final optical loss, as long as the threshold condition is met. The loss threshold for bleeding detection can be appropriately tuned as per the endoscopist’s preference based on the clinical requirements of the procedure, as well as specific conditions of the individual patient.

2.3.1. Finite Element Analysis of Sleeve and Colonoscope Stiffness

The integration of the soft blood sensor sleeve could potentially alter the stiffness properties of the colonoscope. It is vital that the sensor sleeve is sufficiently flexible to avoid excessively stiffening the colonoscope. This will prevent any impediments to the navigation of the colonoscope through tortuous bends i.e., in the sigmoid colon. Thus, the sensor sleeve and the colonoscope were modelled in Abaqus FEA (Simulia, Dassault Systems) to evaluate their stiffness properties. The colonoscope and plastic base were considered as linear elastic materials while the flexible resin sensor sleeve was modelled as a hyper elastic material (see Section 4.4). The sensor sleeve was constrained to the center of the colonoscope and the proximal tip of the colonoscope was constrained against any motion in the X, Y, and Z axes. Two scenarios were simulated to ascertain the impact of the sensor sleeve on colonoscope stiffness. The first scenario consisted of only the colonoscope subject to a distal tip displacement of 10 mm. The second scenario simulated the colonoscope and sensor sleeve attached together, subject to an axial displacement of the same magnitude at the tip.

As seen in Figure 3A–B, the simulation scenarios with and without the sensor sleeve have similar stress distributions and magnitudes. In both simulation scenarios, the average stress is highest in the regions of the colonoscope near the constrained proximal tip. The magnitude of stress decreases in regions of the colonoscope that are further away from the fixed proximal tip. The maximum magnitude of the stress distribution with the sensor sleeve attached is marginally higher than the no sensor case, indicating that the sensor sleeve slightly stiffens the colonoscope. The maximum Mises stress encountered with the sensor attached 0.315 MPa compared to the 0.247 MPa without the sensor on the colonoscope. This is verified in Figure 3C, which shows that the force experienced by a node at the tip of the colonoscope is slightly higher in the simulation case with the sensor sleeve attached. The force reaches a maximum value of 0.61 N when the sleeve is attached onto the colonoscope, whereas the max force is 0.475 N when the sensor sleeve is not attached.

3. Finite element modeling results.

A) Von Mises stress distribution of colonoscope subject to 10 mm tip displacement. B) Von Mises stress distribution of colonoscope and sensor sleeve subject to 10 mm tip displacement. C) Force vs displacement graph of selected node at the colonoscope distal tip. D) Colonoscope subject to 10 mm tip displacement. E) Colonoscope and sensor sleeve subject to 10 mm tip displacement. F) Force vs displacement graph of the colonoscope with and without sensor sleeve. The solid line is the mean value and the shaded area represents the standard deviation computed on three prototypes.

2.3.2. Colonoscope Stiffness Validation

The influence of the sensor sleeve on the colonoscope was experimentally validated through a force versus displacement test. The colonoscope (EC-760R-V/L, Fujifilm) was clamped to a test fixture and subject to tip displacement of 10 mm using a 3D printed indenter mounted onto an Instron machine. Subsequently, the 3D printed flexible resin sleeve was mounted onto the colonoscope and subject to the same displacement (Figure 3D–E). As seen in Figure 3F, the force versus displacement curves for the colonoscope with the sensor sleeve and without the sensor sleeve are similar. The addition of the sleeve slightly increases the stiffness of the colonoscope – requiring an additional average force of 0.07 N compared to when there is no sleeve on the colonoscope. The colonoscope with the sensor sleeve requires an average force of 0.38 N to displace its tip by 10 mm, compared to the colonoscope without the sensor sleeve, which requires an average force of 0.31 N. The difference between the force versus displacement values between the finite elements analysis simulations and the experimental validation could be attributed to the colonoscope material properties. These were based on prior works^38,39 that attempted to characterize the stiffness of other colonoscope models. Thus there may be some degree of variability between colonoscope models (see Section 4.4).

2.4. Incident Force Testing

As the sensor navigates through the colon, it is subject to incident forces and deformations that influence the light transmission across its waveguides. Thus, we subjected the sensor to a tensile compression test to demonstrate the sensor’s ability to filter out loss spikes caused by incident forces. The sensor sleeve was securely mounted onto a test fixture (Figure 4A) and subjected to a compressive displacement of 1.5 mm using a 3D printed indenter attached to an Instron testing machine (5943 Instron, USA). As seen in Figure 4B, the optical loss experienced by the visible and infrared light waveguide show a similar upward trend as the incident force increases. The magnitude of the average infrared loss values are close to those of the visible loss. As the incident force increases from 0 N to 28 N, the average infrared loss increases from 0 to a peak value of 5.6 dB. Similarly, the visible loss increased to a peak value of 4.4 dB. Consequently, when the loss is subtracted as per Equation 3, the net loss is negative and of a significantly lower magnitude. The net loss reaches a peak value of 1.2 dB at a max force of 28 N. Additionally, the net loss ranges between 0 to −1 dB for forces up to 25 N. It is evident that the sensor can consistently filter out loss caused by physical deformations for a large force range. This is a critical performance requirement as peak forces during colonoscopy are highly variable⁴⁰. Thus, the sensor can differentiate signal changes caused by incident force by subtracting optical loss across both waveguides. Since the magnitude of optical loss is similar for both the infrared and visible waveguide, the net loss is below the blood detection threshold. This ensures that physical deformations are not falsely identified as bleeding events.

4. Force testing setup and optical loss behavior of the sensor in response to incident force.

A) Test setup showing blood sensor sleeve mounted on cylindrical fixture underneath the Instron indenter. B) Plot displaying optical loss of the sensor in response to incident force. The solid line is the mean value and the shaded area represents the standard deviation computed on three prototypes.

2.5. Ex vivo Testing Procedure

The performance of the blood sensor was evaluated by endoscopists in an ex vivo test (Figure 5). The study included 10 users, consisting of five novice users and five experts, respectively (see Note S3 and Table S1, Supplemental Information). The triplet of blood sensors were mounted onto the colonoscope while a simulated colonoscopy was performed in an explanted bovine colon (Figure 5B–D). The tissue specimen (Figure 5E–F) was placed into a ColoEASIE-2 Simulator (Endosim LLC) to mimic the anatomical regions of the human colon i.e., cecum, ascending colon, transverse colon, descending colon, sigmoid colon, and rectum. The explanted colon measures approximately 60 cm in length. During the test, the endoscopists perform a complete navigation through the colon in the simulator, passing through the three regions till the distal tip of the colonoscope reaches the end of the model (Figure 5C–D). The time taken for complete navigation is recorded. The ex vivo evaluation consists of five trials with and without the sensors attached on the colonoscope for each user. First, the endoscopist navigates through the colon with only the colonoscope, without the sensor sleeve attached. Then the sensor sleeve is attached onto the colonoscope and the endoscopists insert the colonoscope 10 cm into the simulator. The channels of the sensor are cleaned and evacuated of any fluid and debris and the sensor baseline is calibrated. The endoscopists then navigate through the colon again (Figure 6A). Subsequently, bleeding is simulated behind the distal tip of the endoscope through a perforation in the colon (Figure 6B). The blood is suctioned in by the blood sensors, and causes an optical loss (i.e., a change in voltage relative to the calibrated baseline). The time taken for blood to be detected is subsequently recorded (Figure 6C–D). This test was performed to evaluate the sensor’s capability to successfully detect bleeding in a simulated clinical environment i.e., the explanted colon (see Video S2, Supplemental Information). The test also assesses the sensor’s ability to integrate into the clinical workflow without impeding the endoscopist from performing a full colonoscopy procedure (i.e., from rectum to cecum). The key metrics evaluated in the test are described below.

5. Schematic of the ex vivo test set-up.

A) Clinical workflow of an actual colonoscopy procedure, highlighting the navigational path and distal camera view for the endoscopist. B) Explanted bovine colon arranged in the EndoSim simulator. The colonoscope navigates through the tortuous regions of the colon with the sensor attached on the colonoscope body. C) Bleeding is simulated at a location on the colon behind the distal tip. D) The sensor approaches the location and detects bleeding. E)-F) Images of the EndoSim ColoEASIE 2 simulator with the ex vivo colon inside.

6. Images showing the ex vivo test. Top left: view from colonoscope camera, top right: graphical user interface, bottom right: external tripod view of colon in the simulator, bottom left: borescope camera view of sensor mounted on the colonoscope.

A) The sensor is mounted onto the colonoscope and is inserted into the colon. The navigational path is highlighted in blue. B) The colonoscope tip navigates past the intended bleeding location and blood is injected into the colon. C) The sensor comes into contact with the blood at the bleeding site. The blood is visible from the bore-scope camera behind the sensor. D) The sensor detects the blood as it enters its microchannel. The graphical user interface alerts the endoscopist of the bleeding.

1. The ability of the colonoscope, with the sensor attached, to safely navigate through the colon without any perforation or mucosal damage. This is to ensure that the addition of the device does not adversely interact with the colon tissue and cause resultant damage, which could compromise patient safety. The colon is visually inspected post procedure in order to verify that the sensor did not cause any perforations or mucosal damage. After each navigation, any perforations or mucosal injury is subsequently recorded.

2. Device fatigue or damage. This is to understand the repeatability of the sensor over multiple procedures and determine if the sensors are prone to fatigue and degradation over time. The sensors should exhibit repeatable behavior across multiple trials without its performance degrading due to fatigue and material stress. This is evaluated by a post-procedural inspection of the sensor, identifying any de-lamination of fibers and vacuum tubing. Any subsequent damage to the sensor is recorded.

3. The bleeding detection capability of the sensor. This is to assess whether the sensor is able to successfully detect bleeding once it is simulated at a perforation site along the colon. This metric also quantifies how long the sensor takes to verify the presence of blood in its microchannel once the bleeding is simulated. The sensor must be able to rapidly detect bleeding once it occurs and notify the endoscopist using the GUI.

4. The benefit of using the sensor measured as the time required to complete the colonoscopy. This is to assess how the integration of the sensor sleeve onto the colonoscope affects the navigational ability of the instrument. Ideally, the sensor should minimally impede the endoscopist from maneuvering the colonoscope through the tight curvatures of the colon and allow them to complete the procedure without significantly prolonging the time required. For each test user, the time to complete the colonoscopy procedure was measured with and without the sensor integrated onto the colonoscope.

5. The learning curve of the users measured as the reduction in time taken to perform the colonoscopy with the device attached across five trials. This is to understand if the addition of the sensor to the colonoscope requires a greater learning or adjustment period for the user to comfortably navigate and perform the procedure compared the control case (no sensor on colonoscope). This is clinically important as the sensor must be able to integrate into the clinical workflow in a manner that does not require significant learning and adaption from the endoscopist compared to how they normally perform the colonoscopy.

6. Mental and physical workload of the procedure measured by the NASA task load index (TLX). This provides a quantitative evaluation of the workload of the colonoscopy when the sensor is attached versus the control scenario. Ideally, the addition of the sensor onto the colonoscope must not significantly add to the mental and physical workload of the endoscopist.

2.6. Ex vivo Sensor Performance

This section covers the sensors performance during the ex vivo evaluation. The subsections herein highlight the sensor’s performance with respect to each of the six metrics outlined prior. Note S4 and Table S2 in the Supplemental Information provide a summary of the sensor’s performance with respect to each of the key metrics.

2.6.1. Safe Navigation

It is vital to ensure the integration of the sensor onto the colonoscope does not increase the risk of colon perforation of tissue and mucosal damage. The sensor demonstrates the ability to safely navigate through the colon without significant perforation or mucosal damage to the tissue. Out of a total of 50 trials, the colonoscope only perforates the colon twice (see Table S2, Supplemental Information). In this case, it was not the sensor that perforated the colon but the distal tip of the colonoscope. This was primarily due to colon simulator set-up which tightly fastens the entrance of the colon to the plastic EndoSim simulator box. This makes the colon taught when the colonoscope presses on the inner lumen during tight bends thereby increasing the propensity of perforation. Thus, the sensor was able to safely navigate the colon set up 96 % of the time, and any perforations were due to the nature of the colonoscope itself as opposed to the sensor.

2.6.2. Device Fatigue and Damage

In our prior work³⁶, the blood sensor was susceptible to detachment of optical fibers from the waveguide interface due to deformations induced by navigation. This affected the quality of data recorded by the sensor and rendered it difficult to navigate without irreparable damage. Thus, the materials used to fabricate the current version of the blood sensor, presented in this paper, were chosen to improve its durability (see Section 4.1). The sensor exhibits low fatigue and resistance to damage from repeated navigation. Out of 50 trials, the sensors only experienced permanent damage in one trial (see Table S2, Supplemental Information). In this case, the body of the sensor did not suffer damage, but the glass optical fiber connected to the sensor broke under tension as the colonoscope navigated a curvature. This rendered the sensor unusable for further trials. Thus the robustness of the sensor is only limited by the fragility of the commercial optical apparatus (i.e glass fibers). In two other trials, the sensor experienced detachment of the vacuum tubing. However, this was easily fixed by reinserting and securing the tube into the sensor cladding.

2.6.3. Bleeding Detection Capability

Over the course of 50 trials, the blood sensor was able to detect the presence of blood in an average time of 5.44 s (see Table S2, Supplemental Information). This time includes a buffer period where the sensor samples the data for multiple consecutive points that cross the loss threshold (2 dB). The sensor is able to detect bleeding 96 % of the time. This data is presented in Table 1. The colonoscope navigation prior to reaching the bleeding location is considered a control (true negative case), i.e., there is no bleeding simulated and the sensor’s signal is observed to ensure there are no false positives due to physical deformation. When bleeding is simulated, the sensor only fails to identify blood in two out of 50 trials (false negative). This is due to error in the sensor calibration, where blood from previous trials accumulated in the microfluidic channel when the signal baseline was measured right before starting the test. However, this is not a significant issue during an actual colonoscopy, as each sensor is designed as a single use device for each colonoscopy. Additionally, the sensor does not record any false positives due to deformation prior to simulated bleeding.

Table 1: Sensor Blood Detection Summary.

Blood sensor confusion matrix displaying bleeding and no bleeding cases (Positive/Negative) and corresponding sensor state (Positive/Negative).

Sensor Status	Bleeding (Positive)	No Bleeding (Negative)
Blood Detected (Positive)	48/50	0/50
No Blood Detected (Negative)	2/50	50/50

As seen in Figure 7, the sensors are able to identify bleeding as the optical loss shows a sustained spike. The sensor alerts the endoscopist after it samples multiple, consecutive data points (n > 3) that are above the set loss threshold of 2 dB. In Figure 7A, the three sensors exhibit minimal loss between time t = 0 s to 80 s. Once bleeding is simulated beyond 80 s, sensor 3 shows a steep rise in loss beyond the threshold, reaching 4.5 dB, and the user is alerted to the bleeding event. In Figure 7B, the user navigates through the colon between 0 s to 75 s. The sensor shows minor spikes below the detection threshold due to air bubbles in the fluid flow through the opto-fluidic interface. Once bleeding is simulated at t = 75 s, sensor 1 and 2 show sustained spikes in loss (2.5 to 3 dB) above the threshold, indicating the presence of blood. Similar behavior is displayed in Figure 7C, the sensors show minimal or zero loss until bleeding is simulated at t = 47 s. Subsequently, sensor 1 shows a spike in loss of magnitude 3 dB while sensor 2 also shows a loss spike above 4 dB. These two sensors indicate the detection of blood due to the simulated bleeding event. The sensor’s ability to filter out false positives from deformation is seen in Figure 7D–E. Figure 7D displays the visible and infrared loss signals for a sensor during colonic navigation prior to bleeding. As seen in the figure, deformation causes spikes in loss across both signals, this is seen specifically at time t = 80 s. However, when the losses are normalized (subtracted), the net loss is close to zero and below the detection threshold (Figure 7E). When bleeding is simulated, the visible loss shows a large rise in signal which leads to the net loss exceeding the threshold and the sensor accurately identifies the bleeding. Subsequent spikes in both visible and infrared loss after bleeding are also normalized at t = 134 s and t = 151 s, thereby preventing false positives.

7. Ex vivo test results displaying sensor loss from four different testing sessions. The graphs display loss data from the sensor triplet as the colonoscope navigates through the colon. When bleeding is simulated at the perforation site, the loss data exhibits sharp spikes above the threshold indicating the detection of blood.

A) Test data from user 1, trial number 4 displaying bleeding detection by sensor 3. B) Test data from user 4, trial 5 displaying bleeding detection by sensor 1 and 2. C) Test data from user 8, trial 2 displaying bleeding detection by sensors 1 and 2. D)-E) Test data from user 10, trial 2 displaying loss of similar magnitude across visible and infrared signals while the sensor navigates through the colon and experiences incident deformation. The net loss from deformation normalizes to a low magnitude below the threshold. The net loss only shows a spike when bleeding is simulated.

2.6.4. Ex Vivo Learning Curve

The learning curves of the endoscopists were measured to evaluate the rate of progress to accomplish the clinical task (i.e., colonoscopic navigation) with and without the blood sensor attached. Ideally, the sensor must integrate into the clinical workflow without requiring a steep learning curve for the user subjects (i.e., the endoscopists). As seen in Figure 8A, the average navigation times for the trials with and without the sensors are similar to each other. The average navigation time without the sensor is 54.8 ± 29.5 s (median: 38.5 s, IQR: 23 s to 70 s). The average navigation time with the sensor is 57.4 ± 19.9 s (median: 48.5 s, IQR: 34 s to 69 s). As seen Figure 8B, the learning curve for the sensor and no sensor trials show a similar downward trend in average navigation time. As the number of trials increases, the navigation time for the sensor case decreases from an initial average value of 78.6 s to an average of 46.5 s by trial 5. Similarly, the average navigation time for the no sensor case decreased from an average time of 79.6 s to 43.3 s by trial 5. The sensor is able to integrate into the clinical task without significantly increasing procedure time or requiring the endoscopists to acquire new skills.

8. Average navigation times and learning curves for users.

A) Average and median navigation time with and without the sensor attached to the colonoscope. B) Average learning curve for all users. C) Average learning curve for experts. D) Average learning curve for novices.

It is worth noting that the average navigation learning curve without the sensor attached to colonoscope has a larger standard deviation due to the novice users. Novice users display large variation in navigation time due to their relative inexperience. Thus, they take longer to perform the task when first introduced to the test set up. The standard deviation of navigation time reduces once the novices test the colonoscope with the sensor attached, as they have already become accustomed to using the colonoscope in the experimental set-up. The expert learning curve shows low standard deviation for both the sensor and no sensor trials. The experts exhibit a consistent downward trend of navigation time with respect to the number of trials. Particularly, they show a large reduction in navigation time between the first and second trials, due to their relative experience with the clinical task. The presence of the sensor on the colonoscope only marginally increases the time required to complete the colonoscopy by the experts. The experts complete the colonoscopy with the sleeve in an average time of 45.0 ± 24.1 s, while they complete the colonoscopy without the sensor in an average time of 25.4 ± 12.3 s. For novice users, the average time to complete the colonoscopy gradually decreases from 117 s to about 68.2 s by trial 5 in the no sensor case. With the sensor attached, the average time decreases from 83.2 s to 55.8 s by trial time. For novices, the average navigation time across 5 trials with the sensor is 69.8 ± 28.1 s. The average navigation time without the sensor is 84.1 ± 44.3 s. This demonstrates that the sensor can be added onto the colonoscope without hindering the learning rate of novices and experts performing colonoscopies.

2.6.5. NASA Task Load Index

A NASA Task Load Index (TLX)⁴¹ evaluation was conducted to assess the performance of the soft bleeding sensor during a colonoscopy with our user subjects. The NASA TLX is a method of assessing the cumulative human workload of specific tasks. The TLX determines the cumulative workload based on six sub-scale ratings including Mental Demand (MD), Physical Demand (PD), Temporal Demand (TD), Effort (E), Performance (P), and Frustration (F). A higher rating for the sub-scales means that the task induced high discomfort and required a large workload for the subject. Overall, a low TLX score signifies an easier workload with low discomfort, implying that user performed a task well according to their self-evaluation. For this evaluation, the users performed two types of navigation tasks through the ex vivo colon with and without the blood sensor attached onto the colonoscope. After each task, the user conducted a pair-wise comparison of the six sub-scales to weight the scales relative to each other. The users also rated each sub-scale from 0 to 100 based on how much each sub-scale contributed to the task workload. The raw rating of the sub-scales is multiplied by the scale weights to obtain the weighted task ratings. The weighted task ratings of the sub-scales are added up and divided by sum of the weights to get the average workload score of the task.

As seen in Figure 9A, the TLX scores are similar for the test cases with and without the sensor attached to the colonoscope. The TLX score for the test case with the sensor is 46.6 ± 14.2 (median: 48.2, IQR: 26 to 65.3), which is slightly higher than the TLX score for the no sensor test case of 43.2 ± 14.4 (median: 43.9, IQR: 24 to 64). The TLX scores for both cases are within the standard deviation range of each other. Expert users exhibit a low TLX workload for both cases as seen Figure 9B. The sensor case for experts has a score of 26.1 ± 5.9 (median: 26, IQR: 22.1 to 28.6) compared to the no sensor score of 23.5 ± 9.4 (median: 24, IQR: 14.1 to 32.1). The sensor is able to integrate into the clinical workflow of the colonoscopy without significantly adding to the task workload. Similarly, the TLX scores for the novices (Figure 9C) show only a marginally higher workload of 66.6 ± 7.8 (median: 65.3, IQR: 59.8 to 72.8) for the sensor case compared to the no sensor workload of 62.9 ± 8.2 (median: 64, IQR: 54.8 to 71.5). For both cases, the highest contributing factors to the workload were mental demand and effort (Figure 9D). The sensor case had higher frustration and physical demand scores than the no sensor case, mainly due to the optical fibers and tubing attached to the sensor. This slightly impeded the insertion of the colonoscope into the simulator box and added friction during navigation. However, the sensor case did have lower (better) performance scores and lower temporal demand scores compared to the no sensor case. The navigation trials with the sensor were performed after the navigation with no sensor (control). This shows that users were able to improve and adapt to the performance and time constraints of the clinical procedure even after the addition of the sensor to the colonoscope. Overall, the blood sensor enables critical bleeding detection capabilities beyond the distal tip of the colonoscope without significantly adding to the workload of the colonoscope, as evidenced by the TLX scores.

9. NASA Task Load Index scores from users with and without the sensor attached onto the colonoscope.

A) Average TLX workload score for all users. B) Average TLX workload score for expert users. C) Average TLX workload score for novice users. D) Average weighted sub-scale ratings contributing to the workload for the two tasks.

3. Discussion

This work presents a novel approach for blood detection in colonoscopy that provides detection capabilities beyond the distal tip of the colonoscope. The sensor novel design optimizes the performance of the device in the ex vivo tissue environment. The sensor’s waveguides utilize visible and infrared light to filter out signal changes caused by incident forces. This also allows the sensor to identify blood through a unique spectral response wherein the visible light is attenuated and the infrared light is minimally attenuated. Thus, the sensor’s optical design allows accurate bleeding detection, minimizing any changes to optical loss caused by other physical phenomena.

The stiffness of the colonoscope integrated with the sensor sleeve was characterized using finite element analysis and experimental validation. It was shown that the sensor sleeve only marginally increases the stiffness of the colonoscope. This ensures that the colonoscope is able to navigate tortuous regions of the colon without significant impediment. This is also practically demonstrated in our ex vivo tests.

The soft bleeding sensor was evaluated in an ex vivo colon by endoscopists of various skill levels, to ascertain the clinical performance of the sensor. The sensor was able to integrate smoothly into the clinical workflow and quickly detects bleeding in an average time of 5.44 s. Additionally, the sensor was able to successfully detect bleeding 96 % of the time across 50 trials. The sensor also demonstrated to integrate onto the colonoscope and navigate safely. Additionally, the sensor exhibited robustness across multiple trials, sustaining irreparable damage only in one trial.

Based on the user data, the blood sensor smoothly integrates into the clinical procedure, without adding to the navigation time, learning curve, and cumulative workload. This is proven by the similar NASA TLX workload scores for colonoscopy with and without the sensor. The results from the ex vivo evaluation prove that the sensor can be easily incorporated into the clinical workflow and provide crucial bleeding detection capabilities behind the distal tip of the colonoscope. This is clinically significant as the results indicate that the blood sensor could effectively serve as an “add-on” device in actual colonoscopy procedures in the future.

Future work will focus on the evaluation of the sensor in a live, in vivo animal model. This will allow us to further assess and demonstrate the clinical benefit of the blood sensor during an actual colonoscopy with living tissue. Additionally, future iterations of the sensor could utilize wireless optoelectronic components instead of optical fibers. This could minimize the complexity of tubing associated with device set-up, as well as reduce potential friction during navigation in tortuous paths. This would also help reducing interference from tubing and wiring when the endoscopist maneuvers the colonoscope.

4. Experimental Procedures

Resource Availability

Lead Contact

Further information for resources and data will be fulfilled by the lead contact, Sheila Russo (russos@bu.edu).

Materials Availability

This study did not generate new materials.

Data Availability

Additional data related to the information presented in the main text and Supplemental Information is available from the authors upon request. This paper does not report original code.

4.1. Fabrication

The whole sensor fabrication process is illustrated in Figure 10. The sensor cladding is manufactured by casting silicone polymer onto silicon wafers (see Video S3, Supplemental Information). Two silicon wafers are used for the cladding fabrication process, one wafer contains patterned features of the microfludic channels and waveguides, while the second wafer is a blank substrate. The patterned wafer is fabricated through UV photolithography (Figure 10A) with a SU-8 2100 photoresist (Kayaku Advanced Materials, Inc.). Polydimethylsiloxane silicone polymer (10:1) is dispensed onto the silicon wafers and spin coated. The wafers are spun at a rate of 150 rpm for a duration of 45 s with a 5 s ramp. This produces a layer thickness of 0.5 mm (Figure 10B). The patterned and blank claddings are heat cured at 80° for 25 min. The PDMS layers are peeled of the wafer and subject to oxygen plasma surface treatment (Figure 10C) using a plasma cleaner (Diener Electronic GmBh, Germany.). The recipe for the procedure is as follows: 70 W, O₂ gas, 0.4 mbar for 35 s. The layers are brought into contact and heated at 70°C for 2.5 min. This forms an irreversible bond between the two PDMS claddings and seals the microchannels. The inlet holes above the empty waveguide channels are created using a 1.5 mm biopsy punch. A high refractive index, UV curable adhesive (NOA 73, Norland Inc.) is dispensed onto the inlet and the sensor is placed in a vacuum chamber at −100 kPa for 10 min. Upon re-pressurization, the pressure differential forces the liquid adhesive into the channels to fill the waveguide core (Figure 10D). Optical fibers are then inserted into the ends of the filled waveguides and the sensor is subject to a UV cure to solidify the optical adhesive (Figure 10E). The optical adhesive bonds onto the glass fiber, thereby minimizing coupling losses induced by movement of the fiber and or de-lamination. The sensor is then placed into a 3D printed mold that aligns the optical fibers at the base of the sensor and Dragon Skin 30 polymer (Smooth-On Inc.) is cast onto the fibers and cured to form an end-cap (70°, 30 min). This provides additional reinforcement to reduce the risk of optical fiber detachment from the sensor during navigation in the colon. The sensor is adhered onto a semi-flexible glass-reinforced epoxy base (FR-4, McMaster-Carr, USA) using a silicone adhesive (Sil-Poxy, Smooth-On Inc.). The sensor is then attached onto a 3D printed flexible resin sleeve (Flexible 80A resin, Form 2, Formlabs Inc.) which allows the sensors to be radially mounted onto the colonoscope (Figure 10F). Tubing is then inserted and secured into the end of the microchannel to connect the sensor to the vacuum and flushing system.

10. Sensor Manufacturing Process.

A) Silicon wafers are patterned using UV lithography. B) PDMS is spin coated onto patterned and blank wafers and then heat cured to form the cladding. C) Patterned cladding is bonded to a blank substrate via oxygen plasma modification. D) Waveguides channels are filled under vacuum with NOA 73 UV adhesive. E) Optical fibers are inserted into the waveguide channels and the core material is UV cured. F) Sensor is bonded onto a 3D printed flexible sleeve that can be mounted onto a colonoscope.

4.2. Sensor Control System

The control system for the sensor consists of an optoelectronic circuit to process voltage signals from the sensor and a pneumatic circuit to control the vacuum and flushing of the microfluidic channels (Figure 11). The waveguides of the blood sensor interface with 200 μm optical fibers (C03597-10-BL, OFS Fitel, LLC). The optical fibers are coupled to LEDs and photo-transistors on the optoelectronic circuit. The circuit consists of 2 types of LEDs, one transmitting green light with a peak wavelength of 530 nm (IF-E93, Industrial Fiber Optics), and another LED transmitting infrared light with a wavelength of 930 nm (IF-E91A, Industrial Fiber Optics). The photo-transistor (IF-D92, Industrial Fiber Optics) is connected to a voltage divider circuit with a 470 kΩ resistor in series. The signal output from the photo-transistors are read by an Arduino Nano micro-controller. The data from the sensors are processed and displayed via a graphical user interface using MATLAB (Mathworks, Inc.). The pneumatic circuit consists of two peristaltic liquid pumps (Adafruit), one pump provides a vacuum suction while the other pump enables flushing of the sensor’s channels. The peristaltic pumps are controlled via a motor driver (DRV8833, Adafruit). The pumps are connected to three solenoid valves (DFR0866, DFRobot). The solenoid valves are controlled using a MOSFET board (Mosfetti, Monk Makes), which allows sequential vacuuming for each sensor connected to the circuit.

11. Blood sensor control system schematic.

Diagram displaying the pneumatic control system (left) and optoelectronic circuit (right).

4.3. Blood Concentration vs Sensor Loss Test Procedure

The blood sensor was evaluated with varying concentrations of the synthetic blood analog (Carolina Biological, USA). For the test, 2 ml samples were prepared by diluted the blood analog with deionized (DI) water using a centrifugal mixer (ARE-310, Thinky Corp.). The concentrations of the blood analog ranged from 20% to 100% volume by volume in 20% increments. A total of four devices were evaluated with five trials for each blood concentration. The sensor baseline was recorded prior to testing the sensor with blood and the baseline was used to calculate the optical loss. This was done prior to testing as the blood analog stains the sensor and gradually reduces the baseline over time. This can result in an inaccurate optical loss value. Additionally, the sensor microchannel was flushed with DI water and then evacuated using a syringe between varying concentrations of blood to avoid mixing.

4.4. Finite Element Analysis Method

The colonoscope and sensor sleeve were modelled in Abaqus FEA (Simulia, Dassault Systems). The colonoscope was considered a linear elastic material with properties were as follows^38,39,42: Young’s Modulus E = 16 MPa, Poisson’s ratio v = 0.3, and a mass density ρ = 2944kg/m³. The properties of the flexible resin sensor sleeve were modelled based on a 2 parameter Mooney-Rivlin hyper-elastic model⁴³. The coefficients are as follows: C₁₀ = 0.62691 and C₀₁ = 0.36902. The colonoscope was meshed with a linear geometric order, reduced integration hex element type that resulted in C3D8R elements. The flexible resin sleeve was meshed with a quadratic geometric order, hybrid hex element type that resulted in C3D20H elements. The glass reinforced epoxy base (FR-4) on the resin sleeve was modelled as a linear elastic material with the following properties^44,45: E = 26000 MPa, Poisson’s ratio v = 0.26, and a mass density ρ = 1940kg/m³. The universal seed size for both the resin sleeve and colonoscope meshes was 1 mm, while the seed size for the FR-4 base was 1.1 mm. The sensor sleeve was tie constrained to the colonoscope and the colonoscope was fixed at the proximal end against translation in the X, Y, and Z axes. A reference point node was selected on the distal tip of the colonoscope and the reference point was coupling constrained to the face of the colonoscope in 6 degrees of freedom. This ensure the entire tip of the colonoscope will be displaced by 10 mm. Two simulation scenarios were evaluated, which involved displacing the colonoscope with and without the sensor sleeve. The force and nodal displacement data were extracted and plotted for the corresponding reference point.

4.5. Colonoscope Stiffness Validation Procedure

The stiffness of the colonoscope with and without the sensor sleeve was experimentally validated using an Instron universal testing system (5943, Instron).The colonoscope was radially clamped to a 3D printed fixture (Grey V4 Resin, Form2, Formlabs Inc.) at a length of 10 cm behind the distal tip of the colonoscope. A 3D printed indenter (Grey V4 Resin, Form2, Formlabs Inc.) was attached to the 50 N load cell of the instron. The indenter is flat at the bottom and measures 3 mm by 40 mm. The colonoscope is subject to a compressive tip displacement of 10 mm at a rate of 1 mm/s and the corresponding reaction force is measured. Similarly, the sensor sleeve is then mounted on the colonoscope and subject to the same test procedure.

4.6. Incident Force Testing Procedure

The sensor response to incident force was experimentally evaluated using using an Instron universal testing system (5943, Instron). The sensor was mounted onto cylindrical 3D printed fixture (Grey V4 Resin, Form2, Formlabs Inc.) measuring 12 mm diameter and 100 mm in length. A 3D printed indenter (Clear V4 Resin, Form2, Formlabs Inc.) was attached to the 50 N load cell of the instron. The indenter is flat at the bottom and measures 3 mm by 40 mm. The indenter is aligned across the center of the sensor and then the sensor is subject to a compressive tip displacement of 1.5 mm at a rate of 1 mm/s. The corresponding sensor output (voltage) from the visible and infrared waveguides is measured through the opto-electronic circuit, along with the reaction force in response to the compressive displacement.

4.7. Optical Modeling Method

The optical lens was first modeled as a sequential system using Zemax OpticStudio. Subsequently the cladding and waveguides were modeled around the lens using the part designer tool and the system was converted into non-sequential mode. The optical properties of the NOA 73 waveguide core material was ascertained by fitting a 5 term Schott dispersion formula to the refractive index data at specific UV-Visible wavelengths. The optical properties of PDMS were ascertained using a 3-term Sellmeier equation⁴⁶ with B₁ = −2.30744, B₂ = 1.85952, B₃ = 1.39944, C₁ = 0.02605, C₂ = 0.01362, and C₃ = 0.01349. The incident light source was set as an entrance pupil diameter aperture type with an aperture value of 0.2. The incident wavelength was set to 540 nm and the spot diagram ray density was set as 10,000 rays for all simulations. The location of the lens was placed behind the microchannel, receiving incident light from the emitting waveguide. Initial dimensions and radial curvature of the lens were approximated based on the optical length space constraint, which was <= 700 μm. The radii and distance between the two curved lens surfaces were then optimized using the quick adjust function with the primary criterion set to minimize the spot size radial (see Note S1, Supplemental Information).

Supplementary Material

2

1. Video S1: Sensor demonstration. Demonstration of sensor’s ability to alert the surgeon of the presence of blood through a graphical user interface (GUI). As the blood analog enters the microfluidic channel of the sensor, the incident green light is attenuated. This causes a spike in loss and the GUI reads “Blood Sensed”.

3

2. Video S2: Ex vivo test procedure. Ex vivo testing setup of the explanted bovine colon placed in the Endosim ColoEASIE-2 simulator. The top left view shows the colonoscope camera video feed, the top right view shows the graphical user interface data, the bottom right view displays the top-down external camera view of the simulator, and the bottom left view shows the borescope camera view of the sensor mounted on the colonoscope. The colonoscope with the sensor navigates through the colon to the bleeding location. Bleeding is simulated by injecting the analog into the colon. A pool of blood is visible in the borescope camera view and the sensor suctions in the blood. The sensor detects the bleeding and the GUI alerts the user that blood is sensed.

4

3. Video S3: Sensor fabrication. Blood sensor fabrication process showing the spin coating of the sensor cladding, plasma surface modification, waveguide core vacuum filling and UV curing, pouring and curing the silicone end cap on the sensor’s fibers, and finally bonding the sensor to the plastic base and resin sleeve.

Highlights:

• Developed a soft blood sensor for colonoscopy

• Sensor provides blood detection behind the colonoscope’s distal tip camera

• Utilizes multiple wavelengths to identify blood real time

• Sensor is evaluated in an ex vivo bovine colon with a cohort of endoscopists

Bigger Picture:

Colonoscopies are essential to diagnose various gastrointestinal diseases. However, the flexibility of the colonoscope often results in the device looping in the colon and applying excessive forces on the colon wall. This can cause severe adverse events, including bleeding. The colonoscope only has visualization at the distal tip camera which can impede accurate identification of bleeding behind the distal tip. Current solutions for identifying bleeding in colonoscopies are limited and do not integrate easily onto traditional colonoscopies. This work presents a soft optical bleeding sensor that can be mounted onto a colonoscope to provide bleeding detection capabilities behind the distal tip. We evaluate the device in an ex vivo study with expert and novice endoscopists.