Early detection and diagnosis of oral cancer are critical for a better prognosis, but accurate and automatic identification is difficult using the available technologies. Optical coherence tomography (OCT) can be used as diagnostic aid due to the advantages of high resolution and non-invasion. We aim to evaluate deep-learning-based algorithms for OCT images to assist clinicians in oral cancer screening and diagnosis. An OCT data set was first established, including normal mucosa, precancerous lesion, and oral squamous cell carcinoma. Then, three kinds of convolutional neural networks (CNNs) were trained and evaluated by using four metrics (accuracy, precision, sensitivity, and specificity). Moreover, the CNN-based methods were compared against machine learning approaches through the same dataset. The results show the performance of CNNs, with a classification accuracy of up to 96.76%, is better than the machine-learning-based method with an accuracy of 92.52%. Moreover, visualization of lesions in OCT images was performed and the rationality and interpretability of the model for distinguishing different oral tissues were evaluated. It is proved that the automatic identification algorithm of OCT images based on deep learning has the potential to provide decision support for the effective screening and diagnosis of oral cancer.

Oral maxillofacial conditions are highly prevalent and have become significant public health challenges worldwide.^1,2 Given the vital role played by oral maxillofacial regions both functionally and esthetically, false diagnosis and subsequently delayed intervention or mistreatment can impose on patients not only heavy economic burden but also great social psychological stress due to potential malnutrition and disfiguring.^3,4 Reliable diagnosis is crucial to formulating tailored treatments and early intervention to effectively prevent further deterioration and to greatly improve the overall outcome and life qualities of patients.^1,5 Histopathological interrogation of biopsy specimens is still the current gold standard of definitive diagnosis of oral maxillofacial lesions. However, the biopsy procedure, typically guided by an upper endoscope, is prone to sampling errors⁶ since in many cases, abnormal tissue structures emerge from subepithelial locations [e.g., basement membrane (BM) for basal cell carcinoma] in premalignant stages,^7,8 and exhibit barely discernible visual difference on the tissue surface. Standard medical imaging modalities, including x-ray, computed tomography (CT), and ultrasonography, are not well-suited for real-time biopsy guidance due to their high cost, limited imaging resolution, and insufficient detection sensitivity of early transformation.⁹ Therefore, there has been compelling need for a noninvasive, high-resolution, and cost-effective imaging technology that is able to visualize inconspicuous subepithelial microscopic abnormalities over an extended area in vivo and in real time and can be used for routine screening or targeted biopsy guidance.

SignificancePathologies within the tympanic membrane (TM) and middle ear (ME) can lead to hearing loss. Imaging tools available in the hearing clinic for diagnosis and management are limited to visual inspection using the classic otoscope. The otoscopic view is limited to the surface of the TM, especially in diseased ears where the TM is opaque. An integrated optical coherence tomography (OCT) otoscope can provide images of the interior of the TM and ME space as well as an otoscope image. This enables the clinicians to correlate the standard otoscopic view with OCT and then use the new information to improve the diagnostic accuracy and management.AimWe aim to develop an OCT otoscope that can easily be used in the hearing clinic and demonstrate the system in the hearing clinic, identifying relevant image features of various pathologies not apparent in the standard otoscopic view.ApproachWe developed a portable OCT otoscope device featuring an improved field of view and form-factor that can be operated solely by the clinician using an integrated foot pedal to control image acquisition. The device was used to image patients at a hearing clinic.ResultsThe field of view of the imaging system was improved to a 7.4 mm diameter, with lateral and axial resolutions of 38 μm and 33.4 μm, respectively. We developed algorithms to resample the images in Cartesian coordinates after collection in spherical polar coordinates and correct the image aberration. We imaged over 100 patients in the hearing clinic at USC Keck Hospital. Here, we identify some of the pathological features evident in the OCT images and highlight cases in which the OCT image provided clinically relevant information that was not available from traditional otoscopic imaging.ConclusionsThe developed OCT otoscope can readily fit into the hearing clinic workflow and provide new relevant information for diagnosing and managing TM and ME disease.

Background: Skull base tumors can extend into the temporal bone and occasionally even be visible through the tympanic membrane (TM) if they grow into the middle ear cavity. The differential diagnosis of a skull base mass is extensive and ranges from non-tumorous lesions like cholesteatoma to benign tumors like schwannoma and to malignant lesions like metastatic cancer. Optical coherence tomography (OCT) is a noninvasive imaging technique that can image tissue with high resolution in three dimensions, including through structures such as the TM and bone. OCT angiography is also able to assess tissue vascularity. We hypothesized that OCT could help shrink the differential diagnosis in clinic on the day of initial presentation. Specifically, we thought that OCT angiography could help distinguish between highly vascular skull base tumors such as glomus jugulare and other less vascular tumors and middle ear pathologies such as cholesteatoma and schwannoma. Objectives: We sought to determine whether OCT can image through the TM in clinic to distinguish a normal ear from an ear with a mass behind the tympanic membrane. Furthermore, we sought to assess whether OCT angiography can detect vascularity in these masses to help inform the diagnosis. Methods: We designed and built a custom handheld OCT system that can be used like an otoscope in clinic. It is based off a 200 kHz swept-source laser with a center wavelength of 1310 nm and a bandwidth of 39 nm. It provides a 33.4 μm axial and 38 μm lateral resolution. Cross-sectional images of the middle ear space, including OCT angiography, were captured in an academic neurotology clinic. Patients with normal ear exams, glomus tumors, cholesteatomas, and facial nerve schwannoma were imaged. Results: OCT images revealed key structures within the middle ear space, including the TM, ossicles (malleus and incudostapedial joint), chorda tympani, and cochlear promontory. OCT also identified middle ear pathology (using pixel intensity ratio in the middle ear normalized to the TM) when compared with patients with normal ear exams (mean 0.082, n = 6), in all patients with a glomus tumor (mean 0.620, n = 6, p < 0.001), cholesteatoma (mean 0.153, n = 4, p < 0.01), and facial nerve schwannoma (0.573, n = 1). OCT angiography revealed significant vascularity within glomus tumors (mean 1.881, n = 3), but minimal vascularity was found in normal ears (mean 0.615, n = 3, p < 0.05) and ears with cholesteatoma (mean 0.709, n = 3, p < 0.01), as expected. Conclusions: OCT is able to image through the TM and detect middle ear masses. OCT angiography correctly assesses the vascularity within these masses. Thus, OCT permits the clinician to have additional point-of-care data that can help make the correct diagnosis.

SignificanceCholesteatomas, benign tumors that grow in the middle ear, can lead to conductive hearing loss. If not completely removed during surgery, these tumors may regrow. Current imaging technologies struggle to detect residual tumors noninvasively due to limitations in contrast and resolution, often necessitating additional surgery for inspection. To address this, we developed a catheter endoscope capable of being inserted through the Eustachian tube, allowing detailed examination of the middle ear without surgery. Using two-photon polymerization (2PP) technology, we fabricated miniature, side-viewing reflective endoscope objectives. This approach enabled the rapid production of single-element objectives with highly repeatable optical properties, easily adaptable to specific imaging needs.AimWe aim to design, fabricate, and demonstrate a catheter endoscope for optical coherence tomography (OCT) endoscopy of the middle ear via the Eustachian tube.ApproachSide-viewing, reflective lenses were designed in OpticStudio and 3D printed using 2PP followed by sputter coating with gold. Standard metrology techniques were used to verify and optimize the objective’s shape. The optical performance of the catheter endoscopes was measured with a beam profiler. Finally, OCT imaging of the middle ear of a pig via the Eustachian tube was completed using the fully assembled catheter endoscope.ResultsMetrology showed the printed lenses conformed closely to the design. The catheter endoscope’s FWHM spot size had a mean ± standard deviation of 25.3±1.8 μm with a measured working distance of 1.960±0.057 mm. Volumetric OCT images of the middle ear, inner ear, and Eustachian tube were captured in a postmortem pig head using the catheter endoscope.ConclusionsThe 2PP approach is fast and highly repeatable for miniature reflective objective fabrication. OCT catheter endoscopy via the Eustachian tube enabled imaging of the middle ear, Eustachian tube, and surprisingly part of the inner ear.