Using the shortwave infrared to image middle ear pathologies

Significance

Imaging with shortwave infrared (SWIR) light has great potential for visualizing biological structures previously undetectable with visible light. To demonstrate the clinical potential of SWIR imaging, we developed a medical otoscope sensitive to SWIR light. We show that the unique transmission of SWIR light through tissue improves resolution of anatomical structures lying behind thin tissue membranes like the ear drum. We therefore significantly improve imaging of underlying middle ear anatomy. SWIR imaging also allows identification of disease characterized by fluid accumulation, as in the diagnosis of otitis media. With successful diagnosis of otitis media estimated at 51% for US pediatricians, objectifying this diagnosis could curb the antibiotic resistance associated with an estimated 2 million overdiagnoses each year.

Abstract

Visualizing structures deep inside opaque biological tissues is one of the central challenges in biomedical imaging. Optical imaging with visible light provides high resolution and sensitivity; however, scattering and absorption of light by tissue limits the imaging depth to superficial features. Imaging with shortwave infrared light (SWIR, 1–2 μm) shares many advantages of visible imaging, but light scattering in tissue is reduced, providing sufficient optical penetration depth to noninvasively interrogate subsurface tissue features. However, the clinical potential of this approach has been largely unexplored because suitable detectors, until recently, have been either unavailable or cost prohibitive. Here, taking advantage of newly available detector technology, we demonstrate the potential of SWIR light to improve diagnostics through the development of a medical otoscope for determining middle ear pathologies. We show that SWIR otoscopy has the potential to provide valuable diagnostic information complementary to that provided by visible pneumotoscopy. We show that in healthy adult human ears, deeper tissue penetration of SWIR light allows better visualization of middle ear structures through the tympanic membrane, including the ossicular chain, promontory, round window niche, and chorda tympani. In addition, we investigate the potential for detection of middle ear fluid, which has significant implications for diagnosing otitis media, the overdiagnosis of which is a primary factor in increased antibiotic resistance. Middle ear fluid shows strong light absorption between 1,400 and 1,550 nm, enabling straightforward fluid detection in a model using the SWIR otoscope. Moreover, our device is easily translatable to the clinic, as the ergonomics, visual output, and operation are similar to a conventional otoscope.

Optical imaging and spectroscopy are well-established methods of investigating tissues in vivo and have proven valuable for many diagnostic applications (1–3). A biological tissue sample is illuminated with light and the structural composition and function of the sample can be determined from the diffusely reflected or transmitted light. Equipment for optical measurements is often simple and portable, interrogations can be carried out without contacting or perturbing the tissue over a wide field of view, and the resulting data are generally easy to interpret, all of which have made optical imaging a powerful technique in clinical settings (1, 4). Most optical imaging thus far uses visible and/or near infrared (NIR) light between 400 and 1,000 nm, which can be detected with inexpensive silicon-based sensors. In this wavelength region, oxygenated and deoxygenated hemoglobin are the primary physiologically active optical absorbers, and the functional status of tissue can be measured based on their relative concentrations (5–8).

Extending optical measurements into the shortwave infrared (SWIR; 1,000–2,000 nm) offers several advantages over visible and NIR imaging for certain in vivo applications (3, 9–11). The SWIR regime features absorption from tissue constituents such as water (near 1,150, 1,450, and 1,900 nm), lipids (near 1,040, 1,200, 1,400, and 1,700 nm), and collagen (near 1,200 and 1,500 nm) that are more prominent than corresponding features in the visible and NIR regions (12). The enhanced sensitivity to these chromophores enables better characterization of changes in their concentration, with recent spectroscopy-based examples in detecting and monitoring cancerous tissues (13–15), burns (3), and intestinal ischemia (16), distinguishing skin bruises from surrounding tissue (17, 18), and discriminating histologically vulnerable and stable plaques of blood vessels in vivo (19–21). In addition, SWIR light offers greater transmission through biological tissue than visible or NIR light due to decreased scattering of photons (9, 22, 23). SWIR light thus provides sufficient optical penetration depth to noninvasively interrogate changes in subsurface tissue features, whereas most visible and NIR imaging is limited to superficial structures (11, 22).

Despite these advantages, the SWIR regime has thus far been underused in optical imaging, and in particular, medical devices have been mostly limited to exploratory or proof-of-principle in nature. The limited development of SWIR technology is in part due to the cost-prohibitive nature of nonsilicon semiconductor detector arrays and restricted access due to national defense-related policies such as International Traffic in Arms Regulations (ITAR). However, recent technological advances and an increasing supply of SWIR detectors has enabled a price drop of roughly an order of magnitude over the time period from 2010 to 2014 (12). Advances in detector fabrication have also produced high performing sensors with significantly smaller form factors and reduced weight. Based on such a newly available sensor, we show here an application of SWIR light in a medical otoscope for evaluating transtympanic middle ear pathologies.

Visible light-based pneumotoscopes have been the diagnostic workhorse for external auditory canal and middle ear examination for over a century. Using this tool, physicians assess middle ear pathologies based on the appearance of the tympanic membrane, or ear drum, and its mobility against hand-generated pneumatic pressure. The pneumotoscope serves as an initial means of identifying tympanic membrane perforations, causes of conductive hearing loss, and most commonly, otitis media, which refers to a continuum of inflammatory conditions of the middle ear including acute infection. Otitis media is second in frequency only to acute upper respiratory infection as the most common illness diagnosed in US children; at least 80% of children will have experienced one or more episodes of otitis media by the age of 3 y (24–27).

However, otoscopic examinations are known to have subjective interpretations, especially in the hands of inexperienced practitioners (28, 29). Successful diagnosis of otitis media is estimated at 51% for US pediatricians, with overdiagnosis of acute otitis media (false positives) occurring 26% of the time (30–32). The resulting variation in antibiotic therapy has made otitis media a primary factor in increased antibiotic resistance (33–35). On the other hand, failure to diagnose otitis media can lead to long-term hearing impairment, intracranial complications, a delay in language acquisition, or formation of destructive skin growths, known as cholesteatoma, which can only be treated by surgical excision (36–39). The limitations of pneumotoscopy are related in part to the thickness of the tympanic membrane, which despite being semitranslucent, reflects, absorbs, and scatters incident light at the surface, limiting signal penetration to deeper middle ear structures or fluid. A study by Rosenfeld of 135 acute otitis media cases diagnosed by US primary care practitioners found that 40 were false positive, 35 of which had no middle ear effusion (40).

Here we describe the development of an otoscope sensitive to SWIR light for more objective diagnoses of transtympanic middle ear pathologies. A SWIR otoscope maintains the ergonomics, visual output, and small footprint of a conventional otoscope, making it easily translatable to the clinic. Meanwhile, SWIR endogenous contrast provides a unique image that can be applied to the identification of middle ear abnormalities. Deeper tissue penetration is achievable with SWIR light than with visible light, allowing better visualization of middle ear structures through the tympanic membrane. A clearer view of these structures would aid in identifying causes of conductive hearing loss. In addition, strong SWIR light absorption of water provides more evident contrast between the presence and absence of optically dense middle ear effusions, which could aid in the diagnosis of otitis media. Therefore, a SWIR otoscope could complement conventional pneumotoscopic diagnoses by providing access to the inherent optical properties of the middle ear across an extended wavelength range.

Results

Light Attenuation of Human Tympanic Membrane Tissue.

To predict the performance of a SWIR otoscope in vivo, we investigated the optical properties of human tympanic membrane tissue. Spectroscopic characterization verifies decreasing light attenuation with increasing wavelength (Fig. 1A), which is consistent with the inverse power law relationship between wavelength and scattering of photons measured previously for skin (22, 23). The enhanced transmission of SWIR light is advantageous for an otoscope, given that light must pass through the tympanic membrane after reflecting off of the middle ear structures, such as the highly reflective ossicles and the promontory, before being detected. Any absorption or scattering of light by the tympanic membrane inevitably decreases the amount of light that can be delivered to the middle ear structures and further decreases the reflected signal intensity as it traverses back to the detector. However, more importantly, scattering decreases image contrast and resolution as reflected light travels indirectly through the tympanic membrane to the detector (Fig. 1 B and C, SI Results, and Fig. S1).

Fig. 1.

Attenuation of human tympanic membrane tissue. As wavelength increases, light scattering by the tissue decreases, causing attenuation to decline proportional to the inverse power law λ^-0.26 (A, dashed gray line). Absorption by water in the tissue causes attenuation around 1,440 nm and absorption by hemoglobin causes small features at 540 and 575 nm. Noise in the spectrum around 800 nm is due to poor detector sensitivity and grating efficiency in this region under conditions of low light transmission. The spectrum indicates that light reflected by the middle ear ossicles and promontory would scatter less through the tympanic membrane for SWIR light (C) compared with visible light (B), resulting in greater visibility (Fig. S1).

Fig. S1.

Quantification of the effect of light scattering by tissue on imaging resolution and contrast. Analysis with a sector star resolution target indicates that the resolution of the SWIR otoscope is not affected when the target is placed under ∼2 mm of Intralipid tissue phantom (A); the resolution was calculated to be 21 lp/mm both in the absence (Left) and presence (Right) of the phantom. Michelson contrast values decrease from 0.67 to 0.21. Due to the relative pixel size of the sensors, the visible imaging system has much greater resolution (85 lp/mm) when there is no phantom above the target (B, Left). However, when the resolution target is covered by ∼2 mm of phantom, the target can no longer be resolved. Michelson contrast values decline from 0.48 for no phantom to 0.07 for ∼2 mm of phantom.

Absorption beyond 1,300 nm is also evident in the tympanic membrane attenuation spectrum due to water in the tissue. Although this absorption attenuates overall signal, it does not decrease image contrast and resolution. Even considering the absorption by water, the overall transparency of the tympanic membrane is greater at SWIR wavelengths than in the visible.

Noninvasive SWIR Imaging of Human Middle Ear Anatomy.

To take advantage of the decreased light scattering of SWIR light, we designed an SWIR otoscope prototype capable of imaging the middle ear from 900 to 1,700 nm (Fig. 2). Using a 5.0-mm speculum, the device images an ∼10.5-mm-diameter circular field of view with maximum resolution of 45 µm (Fig. S2).

Fig. 2.

Schematic and image of the SWIR otoscope. The SWIR otoscope prototype is composed of a compact InGaAs SWIR detector, a filter holder, a pair of achromatic doublet lenses, a fiber-coupled light source, and a disposable medical speculum (A). The device is 20 × 17 × 5 cm in size and weighs ∼500 g (B).

Fig. S2.

Characterization of the SWIR otoscope performance. Analysis with a concentric square calibration target (A) and sector star resolution targets with 36 (B) and 72 bars (C) indicates that the imaging system can achieve a maximum resolution of 22 lp/mm or 45 μm and demagnifies the object by a factor of 0.76 onto the sensor with minimal aberrations.

We tested in vivo performance of the SWIR otoscope on ten adults (18 ears) and show that the increased penetration of SWIR light through the tympanic membrane enables the SWIR otoscope to image middle ear anatomy with exceptional detail compared with visible otoscopy (Fig. 3, Figs. S3 and S4, and Table S1 summarize the results). In a typical visible examination, the only clearly identifiable features besides the tympanic membrane are those that are large or superficial, such as the cochlear promontory (15/18 cases), which is formed by the outward projection of the first turn of the cochlea against the posterior wall of the middle ear cavity, and the malleus (18/18 cases), which lies directly under the tympanic membrane. In a few cases, the tympanic membrane is thinner, and light reflection off of smaller or deeper anatomy such as the incus (9/18 cases), stapes (2/18 cases), and stapedial tendon (2/18) is identifiable.

Fig. 3.

Schematic of the middle ear and comparison between SWIR and visible examinations in healthy adults. Representative images from examinations of 10 healthy adult humans are shown. Under visible examination, all anatomy besides the malleus is obstructed by the tympanic membrane (B). Using the SWIR otoscope, the chorda tympani, malleus, incus, stapes, stapedial tendon, cochlear promontory, and round window niche (indicated by ct, m, i, s, st, p, and rw, respectively, and shown schematically in A) are all clearly identifiable (C). See Fig. S3 for outlines of the anatomical features and Fig. S4 for additional SWIR images.

Fig. S3.

Outline of the middle ear structures identified in Fig. 3. In this figure the anatomical structures of the middle ear as identified in Fig. 3 have been outlined to make obvious their position. The malleus, incus, stapedial tendon, cochlear promontory, and round window niche are indicated by m, i, st, p, and rw, respectively.

Fig. S4.

SWIR light examination of healthy adult human subjects. Imaging the middle ear with the SWIR otoscope shows good contrast of the malleus, incus, stapes, cochlear promontory, and round window niche (m, i, s, p, rw) in a typical, healthy adult. In some cases, it is also possible to image the stapedial tendon (st), which stabilizes the stapes against the middle ear canal.

Table S1.

Number of ears in which each middle ear anatomical structure was identifiable using visible vs. SWIR otoscopy

Otoscopy wavelength	Malleus	Incus	Stapes	Stapedial tendon	Promontory	Round window	Chorda tympani
Visible	18/18	9/18	2/18	2/18	15/18	8/18	4/18
SWIR	18/18	16/18	11/18	12/18	18/18	16/18	9/18
Z-score	—	−2.5	−3.1	−3.4	−1.8	−2.8	−1.7
P value	—	0.0062	0.00097	0.00034	0.036	0.0026	0.045

A two-proportion z-test was used to assess whether the difference between the visible anatomy visualization and SWIR anatomy visualization is significant. At a significance level of 0.05, the proportion of anatomy that could be visualized using SWIR otoscopy was significant for the incus, stapes, stapedial tendon, cochlear promontory, round window, and chorda tympani. The difference was insignificant for the malleus, which was easily visualized by both techniques.

The SWIR otoscope, on the other hand, images landmarks of the entire ossicular chain (middle ear bones) (Movie S1), including the incus and stapes (16/18 and 11/18 cases, respectively), in addition to the malleus (18/18 cases). In 12 of 18 cases, it was also possible to image the supporting stapedial tendon. Besides the ossicular chain, the SWIR otoscope can also clearly image the cochlear promontory (18/18 cases), and visualization of the round window niche—one of the two openings from the middle ear to the inner ear—was also achieved (16/18 cases). Furthermore, the SWIR otoscope could identify the chorda tympani (9/18 cases), a branch of the facial nerve that carries taste sensation from the anterior two-thirds of the tongue. The location of the chorda tympani is generally obstructed by a thicker region of the tympanic membrane with visible imaging (visible in 4/18 cases).

Thus, the SWIR otoscope enables visualization of anatomical features that would normally be undetectable due to poor transmission of visible light through the tympanic membrane. For those middle ear structures already detectable by visible examination, the SWIR otoscope further improves the contrast of these features; we show that the contrast of the incus is enhanced by two times on average for the round window and four times on average for the incus (Fig. S5 and SI Results).

Fig. S5.

Quantification of contrast in SWIR and visible otoscopic examinations of healthy adult human subjects. The intensity profile was plotted across the middle ear features as imaged using SWIR otoscopy (column 1, orange trace) and visible otoscopy (column 2, blue trace) to calculate the relative contrast of each imaging method. The round window of one subject is shown in A with Weber contrast values inset in the intensity plot (0.57 for SWIR and 0.27 for visible). The incus of a separate subject is shown in B with SWIR contrast 0.87 and visible contrast 0.63, and the incus of a third subject is shown in C with SWIR contrast 0.43 and visible contrast 0.12. The same procedure was carried out and the Weber contrast plotted for eight volunteers in whom the incus was identifiable by both methods (D) and seven volunteers in whom the round window was identifiable (E). In all cases, SWIR otoscopy provides the greatest contrast of the middle ear feature.

Light Attenuation of Middle Ear Fluid.

As otitis media is the most common pathology observed with an otoscope, we investigated the potential of SWIR light to improve the accuracy of this diagnosis. One of the greatest challenges in the diagnosis of otitis media is the reliable identification of fluid accumulation, or effusion, behind the tympanic membrane in the middle ear. When middle ear effusion is accompanied by rapid onset of one or more signs or symptoms of infection, such as fever, pain and/or irritability, the patient is diagnosed with acute otitis media and treatment with antibiotics is indicated. We acquired several human middle ear fluid samples of a variety of consistencies, ranging from serous to mucoid, to assess their optical characteristics and determine at which wavelength the greatest optical contrast can be achieved.

Spectroscopic characterization confirms strong attenuation of specific spectral bands of SWIR light (Fig. 4). Water in the fluid gives rise to absorption observed in small features around 970 and 1,180 nm and strong features around 1,440 nm and beyond 1,800 nm, corresponding to the vibrational overtone of the O-H bond, and the first overtone of O-H stretching, respectively. Thicker-consistency mucoid middle ear fluid samples contain less water relative to thin samples, but water absorption is still the dominant cause of attenuation (Fig. S6). Two peaks are also distinguishable at 540 and 575 nm from the absorption of oxygenated hemoglobin, but middle ear fluid generally appears translucent by eye due to minimal absorption of chromophores at visible wavelengths (see Fig. S7 for variation in hemoglobin content). Light scattering processes that occur within the viscous mucous also cause attenuation, particularly at visible wavelengths of light (400–700 nm). The strength of this wavelength dependence is a complicated function of the geometry of the scattering particles, which vary with the composition of individual fluid samples; however, in general, the attenuation steadily rises with decreasing wavelength. In the thickest samples, attenuation is as strong at visible wavelengths due to scattering, as it is in the SWIR due to water absorption.

Fig. 4.

Optical properties of human middle ear fluid. The attenuation of light through a sample of middle ear fluid shows strong absorption between 1,400 and 1,550 nm due to water content. This absorption of SWIR light causes the fluid to appear black in a SWIR image, whereas it is translucent with visible imaging or by eye.

Fig. S6.

Attenuation of light by serous, centrifuged serous, and mucoid human middle ear fluid. Attenuation of a single serous, or thin, middle ear fluid sample (thin solid line) shows strong absorption between 1,400 and 1,550 nm and beyond 1,800 nm from water present in the fluid. Centrifuging a separate serous fluid sample isolated this absorptive component from the scattering particles (e.g., cells), and attenuation of the supernatant solution (bold solid line) shows the strong water absorption isolated from the majority of the attenuation due to scattering. Attenuation due to light scattering is more evident in a thicker, or mucoid, fluid sample (dashed line) and increases with decreasing wavelength. Absorption between 1,400 and 1,550 nm is less for the thicker sample due to less water; however, water absorption is still the dominant cause of attenuation.

Fig. S7.

Variability in the optical properties of serous middle ear fluid. The appearance and attenuation of two separate serous middle ear fluid samples were compared. Visible images (A and D) show the color change associated with variations in hemoglobin concentration which correspond to the peaks at 540 and 575 nm in the attenuation spectra (C and F). SWIR images taken with a 1,300-nm long-pass filter (B and E) show the absorption due to water which corresponds to the attenuation spectra peaks between 1,400 and 1,550 nm. In both cases, water absorption allows the fluid to be easily detected using SWIR imaging.

We therefore expect SWIR wavelengths to provide better optical contrast than visible and NIR imaging in the detection of middle ear fluid, due to the strong absorption endogenous to water in the fluid, regardless of fluid consistency.

Detection of Middle Ear Fluid Phantom in a Middle Ear Model.

We evaluated the effectiveness of the SWIR otoscope at detecting fluid by imaging fluid in a 3D-printed middle ear model (Fig. 5). In our model, orange juice was selected as a phantom for middle ear fluid, as it has representative spectroscopic properties, particularly the dominant water absorption features and scattering of visible wavelengths of light (Fig. S8). Using visible otoscopy, slow addition of fluid into the ear model is barely perceptible (Movie S2); reflected light intensity from the middle ear space decreases by less than 30% (Fig. S9A). The malleus, which is a superficial anatomical structure, experiences relatively less change in intensity as fluid is added. As a result, the Weber contrast of this feature, defined as the difference between the feature intensity and the background intensity all divided by the background intensity, is doubled from 0.36 to 0.72 when the malleus is surrounded with fluid (Fig. 5C).

Fig. 5.

Contrast between the presence and absence of fluid in a model middle ear with visible vs. SWIR otoscopy. As the model middle ear is filled with fluid phantom, a subtle intensity change behind the membrane can be observed using visible otoscopy (A). The SWIR otoscope provides a more striking contrast between presence and absence of fluid, particularly with a 1,300-nm long-pass filter that selectively passes the wavelengths of maximum fluid absorption (B). Monitoring the Weber contrast of the model structure shows that the contrast of this feature using SWIR otoscopy increases fourfold to 1.2 (D) compared with contrast in the absence of fluid of 0.30. This increase is a significant improvement over the twofold contrast increase from 0.36 to 0.72 observed using visible otoscopy (C). The Weber contrast for each case is plotted in E.

Fig. S8.

Selection of orange juice as an optical phantom for middle ear fluid. The attenuation spectrum of water (thin solid line) shows the same dominant absorption features as serous middle ear fluid (bold solid line). However, water lacks the scattering properties that attenuate visible wavelengths of light in middle ear fluid. Orange juice is a more representative phantom for middle ear fluid than water, as the attenuation (dashed line) includes both water absorption and scattering of visible light.

Fig. S9.

Quantification of the background light reduction in the model middle ear with addition of fluid. Filling the model ear with fluid causes a 29% reduction in mean background light intensity using visible detection (A), whereas the SWIR otoscope with 1300-nm long-pass detection shows a 73% reduction in background light intensity (B) (Fig. 5). Shown in C is the addition of fluid phantom to the model middle ear observed with the SWIR otoscope without a 1300-nm long-pass filter (detecting wavelengths between ∼900 and 1,700 nm). Filling the model ear with fluid causes a 56% reduction in background light intensity (E). The Weber contrast of the feature in the model is 0.34 when there is no fluid and 0.76 when fluid is present (D), an increase by a factor of 2.25, which is between the twofold and fourfold increase observed with visible and 1,300-nm long-pass SWIR otoscopy, respectively. Thus, the broadband SWIR otoscope still provides better contrast of fluid than visible otoscopy but underperforms compared with 1,300-nm long-pass SWIR detection, which selects out the maximum fluid absorption.

In contrast, using a 1,300-nm long-pass filter and SWIR detection, the SWIR otoscope can selectively image between 1,300 and 1,700 nm where absorption of middle ear fluid is maximal. Addition of fluid to the model (Movie S3) obstructs any features lying behind the superficial structures at the tympanic membrane and reduces the overall reflected light intensity by nearly 75% (Fig. 5B and Fig. S9B). This notable intensity reduction causes contrast of the malleus to increase fourfold from 0.30 to 1.2 when surrounded by fluid (Fig. 5D)—a significant improvement over the visible case (Fig. 5E).

Based on these models and spectroscopic characterization, we therefore expect the imaging contrast of middle ear fluid in vivo to be significantly enhanced with SWIR light compared with what is currently available with visible otoscopy. The relative level of SWIR absorption, which provides a striking contrast, could facilitate a clinician’s determination of the absence or presence of fluid in the middle ear.

SI Results

Quantification of Resolution and Contrast in Vitro for SWIR and Visible Otoscopic Examinations.

Using a sector star resolution target and liquid Intralipid tissue phantom (52–54), we assessed the effect of tissue attenuation on resolution of the SWIR otoscope and a visible otoscope analog (Fig. S1). The phantom was prepared by diluting Intralipid 20% (Baxter Healthcare Corporation) to 2% g/mL fat in water. The 2% fat emulsion was added to a glass-bottom microwell dish (MatTek Corporation) between 0.5 and 2 mm deep. The Petri dish was placed on top of the sector star resolution target, and images were acquired in reflection geometry (illumination and detection from above the phantom). The visible otoscope analog was assembled using a Thorlabs 1,280 × 1,024-pixel CMOS color sensor, and a mounted achromatic lens pair (Edmund Optics) with 75 and 100 effective focal length (EFL) achromats (MgF₂-coated for visible wavelengths). The visible camera and lens were adapted to the otoscope head, and broadband illumination from the otoscope was used to illuminate the resolution target through the phantom at three different thicknesses (∼0.5, 1, and 2 mm). The same procedure was carried out with the SWIR otoscope.

It was determined that the resolution of the SWIR otoscope is primarily limited by the pixel size of the sensor, 30 μm, and is not affected by the optical properties of the thin tissue phantom. At a magnification of 0.76, the system is fundamentally limited to a resolution of no less than 39 μm (or no greater than 26 lp/mm). We observed that adding Intralipid to the top of the resolution target had no effect on the resolution of the target below, even at the greatest thickness of phantom (Fig. S1A). This observation was true for both broadband SWIR detection, which had a resolution of 21 lp/mm with and without phantom, and 1,300-nm long-pass detection, which had a resolution of 20 lp/mm with no phantom and 19 lp/mm with a 2-mm phantom. For the visible imaging system, on the other hand, the resolution was strongly dependent on the thickness of the phantom. Because the pixel size of the CMOS camera is much smaller (3.6 μm), the imaging resolution is much higher, even through the phantom, until at some point the transmission of the phantom is too low to observe the resolution target below. Resolution values were 85 lp/mm for no phantom, 53 lp/mm with ∼0.5 mm of phantom, 38 lp/mm with ∼1 mm of phantom, and at 2 mm of phantom, the target could not be resolved.

Michelson contrast values were also calculated for each assembly, defined as

$$\frac{I_{\max }-I_{\min }}{I_{\max }+I_{\min }},$$

with I_max and representing the highest and lowest intensity values, respectively. Contrast values for broadband SWIR detection decrease from 0.67 to 0.21 when the resolution target is covered with ∼2 mm of phantom and from 0.70 to 0.23 for 1,300-nm long-pass SWIR detection. For visible detection, contrast values decline more significantly; starting from a contrast value of 0.48 for no phantom, the contrast decreases to 0.16 with ∼0.5 mm, 0.12 with ∼1 mm, and 0.07 for ∼2 mm of phantom.

Based on these results, we conclude that the strong scattering of visible light by tissue will inhibit signal delivery to and from middle ear anatomy (which must pass through the tympanic membrane in both directions) and will reduce the resolution and contrast of these structures. We expect that the greater transmission of SWIR light through the tympanic membrane will improve the visualization of anatomy behind the thin tympanic membrane.

Human Middle Ear Image Analyses.

Ten adults were imaged in total in both left and right ears. In 18 of 20 ears, clear visualization of the middle ear was achieved; 2 of 20 ears were excluded due to cerumen in the external auditory canal blocking access to the middle ear. For the 18 ears with a clear view of the middle ear, the individual anatomical structures were identified by the otolaryngologist performing the examination (Table S1). Using visible otoscopy, visualization of the malleus was achieved in 18/18 cases, the incus in 9/18 cases, the stapes in 2/18 cases, the stapedial tendon in 2/18 cases, the promontory in 15/18 cases, the round window in 8/18 cases, and the chorda tympani in 4/18 cases. Using the SWIR otoscope, visualization of the malleus was achieved in 18/18 cases, the incus in 16/18 cases, the stapes in 11/18 cases, the stapedial tendon in 12/18 cases, the promontory in 18/18 cases, the round window in 16/18 cases, and the chorda tympani in 9/18 cases.

A two-proportion one-tailed z-test was used to assess whether the difference between the visible anatomy visualization and SWIR anatomy visualization is significant. We hypothesized that the proportion of volunteers in which the middle ear anatomy could be visualized would be greater using SWIR otoscopy than using visible otoscopy. Using a one-tailed test, and at a significance level of 0.05, the proportion of anatomy that could be visualized using SWIR otoscopy was significantly greater for the incus, stapes, stapedial tendon, cochlear promontory, round window, and chorda tympani. The difference was insignificant for the malleus, which was easily visualized by both techniques. At a significance level of 0.01, the difference was significant for visualization of the incus, stapes, stapedial tendon, and round window and insignificant for the malleus, promontory, and chorda tympani.

Quantification of Contrast in Vivo for SWIR and Visible Otoscopic Examinations.

Using ImageJ software, a region of interest line was drawn across the incus and the round window in corresponding locations of SWIR (Fig. S5, column 1) and grayscale visible images (Fig. S5, column 2). This procedure was done for each of the images in which the anatomy was identifiable as indicated above. The intensity profile (Fig. S5, column 3) was plotted across the middle ear features, and the contrast value from each imaging method was calculated from these intensity values. Contrast was quantified using Weber contrast, defined as

$$\frac{I_f-I_b}{I_b},$$

where I_f is the average intensity of the region of interest across the middle ear feature, and I_b is the average intensity of the background region of interest. The absolute value was taken for the round window contrast, because the round window is given by negative contrast relative to the reflected light from the promontory.

This analysis was carried out for the incus and round window of all of the imaged ears. The average contrast value for the incus of all ears was 0.21 using visible otoscopy vs. 0.61 using SWIR otoscopy—an increase by a factor of 4. The average contrast value for the round window of all ears was 0.26 using visible otoscopy vs. 0.52 using SWIR otoscopy—an increase by a factor of 2. Anatomy that could not be identified in the images was excluded from these averages. For those ears in which the anatomy was visible using both SWIR and visible otoscopy, the contrast is plotted and shown in Fig. S5 D and E.

Discussion

We developed an otoscope sensitive to SWIR light for middle ear disease diagnostics. We show that deeper tissue penetration of SWIR light enhances contrast and enables better visualization of middle ear anatomy through the thin tissue of the tympanic membrane. Although the middle ear anatomy is obfuscated by the tympanic membrane during visible light examinations, SWIR otoscopy can be used to examine the ossicular chain, cochlear promontory, round window niche, and chorda tympani. The ability to inspect the ossicular chain in greater detail could provide valuable diagnostic information in cases of conductive hearing loss such as in ossicular discontinuity or otosclerosis, a disorder characterized by abnormal bone growth. Imaging deeper within the middle ear can also allow evaluation of cholesteatoma extension within the middle ear, especially in cases where the ossicular chain is suspected to be involved. The round window is used as an insertion site for cochlear implant electrodes and has also recently been used as an implantation site for hearing aid transducers (41–43). Clear visualization of the round window using a SWIR otoscope could provide an alternative to radiographic imaging for evaluation of such surgical implants. Thus, a SWIR otoscope has diagnostic potential in evaluating the cause of a variety of middle ear complications and in informing surgical procedures.

Furthermore, we predict that a SWIR otoscope can indicate middle ear effusions based on the strong light absorption of middle ear fluid beyond 1,300 nm. Otoscopy is the most widely used technology for assessing middle ear effusions; however, studies have shown that current visible light-based otoscopy is limited in accuracy, with correct interpretation by 46% of general practitioners, 51% of pediatricians, and 76% of otolaryngologists (28, 32, 44). Integrating SWIR light into otoscopy extends the available wavelengths to a regime in which endogenous contrast of middle ear fluid is much greater than at visible wavelengths. It should therefore provide a more objective determination of the presence or absence of middle ear effusion. Bringing objectivity to this diagnosis, which has long been plagued by diagnostic and therapeutic inconsistency, has the potential to reduce overprescription of antibiotics and unnecessary tympanostomy tube surgeries, which is the most common surgical procedure performed in US children (45). Potentially, an SWIR otoscope could also be used for differential detection of mucoid versus serous middle ear effusion, which is also of clinical interest. Measuring the attenuation of the middle ear at different wavelength regimes (e.g., 1,100 vs. 1,450 nm) simultaneously should enable distinction of thin vs. thick ear fluid with high accuracy, possibly giving a measure of the volumetric percentage of water in the fluid.

We underscore the importance of having a photonic method for direct visualization of the functional status of biological tissue. Tympanometry is an audiometric method shown to raise diagnostic success of otitis media to 83%; however, this method is not widely used in primary care settings, largely because nonspecialists lack training in this technique and are inhibited by the cost (28, 35, 46). Emerging technologies such as optical coherence tomography, spectral gradient acoustic reflectometry, and sonography have likewise shown potential for improving the diagnosis of otitis media, but thus far have not been widely adopted in general practice due to the difficulty of data interpretation or unfamiliarity of physicians with their use. (47–50) Using SWIR technology does not add time or complexity to a diagnosis and requires little additional training of medical practitioners who have already been trained in otoscopy. It is thus easily integrated into the clinic.

The general architecture of the SWIR otoscope, which provides immediate functional information, could be extended to the development of a variety of other medical devices to assist in a wide range of surgical procedures throughout the airway and gastrointestinal tract. Other disease conditions characterized by the buildup of fluid could be characterized by the enhanced contrast due to endogenous SWIR absorption. The SWIR otoscope is therefore only an initial example of how extending optical measurements into the SWIR can provide complimentary visual data and address existing limitations of conventional visible light-based medical devices.

Methods

Spectroscopy of Human Tympanic Membrane Tissue and Middle Ear Fluid.

Human tympanic membrane tissue samples (∼1-mm sections) were collected intraoperatively from pediatric patients (ages 0–18 y) during a typical tympanoplasty procedure. Human middle ear fluid samples (30- to 200-μL volumes) were collected intraoperatively from pediatric patients during myringotomy and placement of pressure-equalizing tubes for standard treatment of recurrent otitis media or persistent middle ear effusion. SI Methods provides information on sample collection, storage, and preparation details. Attenuation was measured using a Cary 5000 UV-VIS-NIR spectrophotometer (Varian/Agilent). All biospecimen sample collection and transfer procedures were approved by the Connecticut Children’s Medical Center Institutional Review Board and were from volunteers who provided informed consent. Samples were used and characterized following the procedures approved by the Massachusetts Institute of Technology Committee on the Assessment of Biohazards and Embryonic Stem Cell Research Oversight and Biosafety Program.

SWIR Otoscope Design.

The SWIR otoscope is composed of a fiber-coupled broadband halogen light source, a medical speculum that guides the device into the ear canal (Welch Allyn), speculums sized 2.5, 3.0, 4.0, or 5.0 mm), and a lens system to focus the diffusely reflected light onto an Indium Gallium Arsenide array detector. The lenses used in these experiments were from an Edmund Optics NIR achromatic lens pair with 75- and 100-mm effective focal length achromatic doublet lenses. The detector is a Xenics XS Trigger Indium Gallium Arsenide detector with a 320 × 256 array of 30-μm pixels and 14-bit analog-to-digital conversion resolution. A filter holder in front of the sensor allows easy adaptation with various short-pass, long-pass, and band-pass filters, enabling optimization of the device sensitivity for a variety of applications.

Human Middle Ear Imaging.

The external auditory canal was used as the optical access for SWIR imaging of middle ear structures in both the left and right ears of healthy adults. The speculum was inserted into the canal within ∼2 cm of the tympanic membrane and the middle ear was illuminated with the light source. Reflected light was collected by the optical system either broadband or filtered through a band-pass, long-pass, or short-pass filter. Visible images of the middle ear were taken of each subject in a similar manner (SI Methods). All participants provided informed consent and consent to publish images, and methods were carried out in accordance with the procedures approved by the Massachusetts Institute of Technology Institutional Review Board and Committee on the Use of Humans as Experimental Subjects, and Connecticut Children’s Medical Center Institutional Review Board, and all procedures were in accordance with the Declaration of Helsinki guidelines.

Middle Ear Phantom.

A middle ear phantom was 3D-printed using computer-assisted design software Solid Edge ST7 (Siemens). The shapes and angles were modeled to resemble the normal middle ear anatomical configuration based on previously described anatomical measurements (51). These models were used as the design files for the 3D printing process. A Makerbot Replicator Desktop 3D Printer (Makerbot Industries) was used with polylactic acid (PLA) as the printing filament. To model the semirigid and angled nature of the external auditory canal, a 3D-printed PLA cast was made, and high-performance platinum silicone (Dragon Skin FX-Pro; Smooth On) was used for the canal castings. The silicone was also cast into a thin, translucent sheet ∼0.5 mm thick for the tympanic membrane and to line the inside of the 3D-printed PLA middle ear cavity, representing the mucosa of the cavity. A small hole in the back of the middle ear cavity was used to add fluid via a syringe into the phantom behind the silicone tympanic membrane while recording video using the SWIR or a visible otoscope inserted in the model canal.

SI Methods

Tympanic Membrane Tissue and Middle Ear Fluid Collection, Storage, and Preparation.

Tympanic membrane tissue samples were collected intraoperatively from pediatric patients (ages 0–18 y) during a typical tympanoplasty procedure. A 1-mm section of solid tissue was cut from the tympanic membrane via tympanic membrane perforation and was flash frozen with liquid nitrogen on Telfa. The dissected tissue was stored on the Telfa in a specimen cup at −80 °C for up to 1 wk before being transferred over ∼3 h in a Styrofoam container of dry ice. Tissue was stored in the second location at −80 °C for 1 wk before being thawed to room temperature and mounted intact on a glass slide. Attenuation was measured using a Cary 5000 UV-VIS-NIR spectrophotometer immediately after thawing.

Human middle ear fluid samples were collected intraoperatively from pediatric patients (ages 0–18 y) during myringotomy and placement of pressure-equalizing tubes for standard treatment of recurrent otitis media or persistent middle ear effusion. Following incision on the tympanic membrane with a Beaver blade, a Juhn Tym-Tap (Medtronic Xomed) middle ear fluid aspirator was used to collect 30- to 200-μL samples from either the left or the right ear into Eppendorf tubes. The collected fluid ranged from serous to mucoid in consistency, designated as such based on visual inspection and ability to suction the fluid. One set of samples was centrifuged at 500 rpm for 5 min on a Beckman Coulter Allegra X-12R centrifuge to separate cells from middle ear fluid and subsequently stored at −80 °C. The other set was not centrifuged and was frozen directly at −80 °C. After 5 d, the middle ear fluid samples were transported over ∼3 h in a Styrofoam container of dry ice to a new location where they were again stored at −80 °C. Immediately before spectroscopic characterization, samples were thawed to room temperature and transferred to 1- or 0.2-mm path-length, demountable quartz cuvettes. Attenuation of each sample was measured using a Cary 5000 UV-Vis-NIR spectrophotometer. Images of noncentrifuged middle ear fluid were taken using broadband visible illumination and detection and broadband SWIR illumination and 1,300-nm long-pass detection on a Xenics XS Trigger 320 × 256 Indium Gallium Arsenide SWIR detector.

SWIR Otoscope Performance Characterization.

Resolution and magnification of the otoscope were determined using Thorlabs positive sector star test targets with 36 and 72 bars and a concentric square calibration target (Fig. S2). Depending on the resolution of the optical system, the sector star bars will appear to merge at some radial distance from the center of the target; by measuring this distance, r, the thickness of a line pair can be calculated using the formula for the chord length

$$c=2r*\sin \frac{\theta }{2},$$

where the angle θ is the number of degrees covered by one pair of light and dark bars. The resolution in lp/mm is thus given by 1/c. Using this method, the maximum resolution for the optical system was determined to be 22 lp/mm, or 45 μm. Analysis with a concentric square calibration target indicated that the lenses de-magnify the object onto the sensor by a factor of 0.76, resulting in an ∼10.5-mm-diameter circular field of view while using a 5.0-mm speculum.

Light intensities output from the modified otoscope were also measured. At a distance of 2 cm away from the end of the otoscope speculum, the broadband illumination between 900 and 1,700 nm is 10–30 mW/cm² depending on the output power of the lamp. Illumination specifically between 1,300 and 1,700 nm was measured to be roughly 1–3 mW/cm².

Human Middle Ear Image Acquisition.

All SWIR images were acquired using Xeneth camera control and imaging software. The Xeneth software was used to set the integration time between 500 and 2,000 μs and record the images. Visible images were taken using white LED illumination and recorded with a video otoscope, such as the Horus HD Digital Scope System (JEDMED) or an otoscope with adapted Thorlabs 1,280 × 1,024 complementary metal–oxide–semiconductor (CMOS) color sensor and ThorCam image acquisition software. All image analysis was carried out using ImageJ software (NIH).

Middle Ear Phantom Assembly.

The middle ear phantom was assembled by stretching the silicone tympanic membrane over the opening of the 3D-printed middle ear cavity and affixing the thick silicone external auditory canal. A small hole in the back of the middle ear cavity was used to add fluid via a syringe into the phantom behind the silicone tympanic membrane. SWIR videos of the phantom were captured using the SWIR otoscope, whereas visible videos were taken with a Thorlabs 1,280 × 1,024 CMOS color sensor. Contrast was evaluated using ImageJ software to select a region of interest in the cavity (background) of the ear model and a region of interest across the model middle ear anatomical feature (Fig. 5 C and D and Fig. S9D). The intensity of each region of interest was monitored as fluid was added to the model (Fig. S9 A, B, and E) and Weber contrast values were calculated both in the presence and absence of fluid for visible, broadband SWIR, and 1,300-nm long-pass SWIR examination (Fig. 5E and Fig. S9D).