Paediatric sleep apnea event prediction using nasal air pressure and machine learning.

Sleep-disordered breathing is an important health issue for children. The objective of this study was to develop a machine learning classifier model for the identification of sleep apnea events taken exclusively from nasal air pressure measurements acquired during overnight polysomnography for paediatric patients. A secondary objective of this study was to differentiate site of obstruction exclusively from hypopnea event data using the model. Computer vision classifiers were developed via transfer learning to either normal breathing while asleep, obstructive hypopnea, obstructive apnea or central apnea. A separate model was trained to identify site of obstruction as either adeno-tonsillar or tongue base. In addition, a survey of board-certified and board-eligible sleep physicians was completed to compare clinician versus model classification performance of sleep events, and indicated very good performance of our model relative to human raters. The nasal air pressure sample database available for modelling comprised 417 normal, 266 obstructive hypopnea, 122 obstructive apnea and 131 central apnea events derived from 28 paediatric patients. The four-way classifier achieved a mean prediction accuracy of 70.0% (95% confidence interval [67.1-72.9]). Clinician raters correctly identified sleep events from nasal air pressure tracings 53.8% of the time, whereas the local model was 77.5% accurate. The site of obstruction classifier achieved a mean prediction accuracy of 75.0% (95% confidence interval [68.7-81.3]). Machine learning applied to nasal air pressure tracings is feasible and may exceed the diagnostic performance of expert clinicians. Nasal air pressure tracings of obstructive hypopneas may "encode" information regarding the site of obstruction, which may only be discernable by machine learning.

The predictive power of geographic health care utilization for unintentional fatal fall rates.

BACKGROUND: Falls are the leading cause of fatal and nonfatal injuries among adults over 65 years old. The increase in fall mortality rates is likely multifactorial. With a lack of key drivers identified to explain rising rates of death from falls, accurate predictive modelling can be challenging, hindering evidence-based health resource and policy efforts. The objective of this work is to examine the predictive power of geographic utilization and longitudinal trends in mortality from unintentional falls amongst different demographic and geographic strata. METHODS: This is a nationwide, retrospective cohort study using the United States Centers for Disease Control (CDC) Web-based Injury Statistics Query and Reporting System (WISQARS) database. The exposure was death from an unintentional fall as determined by the CDC. Outcomes included aggregate and trend crude and age-adjusted death rates. Health care utilization, reimbursement, and cost metrics were also compared. RESULTS: Over 2001 to 2018, 465,486 total deaths due to unintentional falls were recorded with crude and age-adjusted rates of 8.42 and 7.76 per 100,000 population respectively. Comparing age-adjusted rates, males had a significantly higher age-adjusted death rate (9.89 vs. 6.17; p <  0.00001), but both male and female annual age-adjusted mortality rates are expected to rise (Male: + 0.25 rate/year, R2= 0.98; Female: + 0.22 rate/year, R2= 0.99). There were significant increases in death rates commensurate with increasing age, with the adults aged 85 years or older having the highest aggregate (201.1 per 100,000) and trending death rates (+ 8.75 deaths per 100,000/year, R2= 0.99). Machine learning algorithms using health care utilization data were accurate in predicting geographic age-adjusted death rates. CONCLUSIONS: Machine learning models have high accuracy in predicting geographic age-adjusted mortality rates from health care utilization data. In the United States from 2001 through 2018, adults aged 85+ years carried the highest death rate from unintentional falls and this rate is forecasted to accelerate.

Predicting Depression From Hearing Loss Using Machine Learning.

OBJECTIVES: Hearing loss is the most common sensory loss in humans and carries an enhanced risk of depression. No prior studies have attempted a contemporary machine learning approach to predict depression using subjective and objective hearing loss predictors. The objective was to deploy supervised machine learning to predict scores on a validated depression scale using subjective and objective audiometric variables and other health determinant predictors. DESIGN: A large predictor set of health determinants from the National Health and Nutrition Examination Survey 2015-2016 database was used to predict adults' scores on a validated instrument to screen for the presence and severity of depression (Patient Health Questionnaire-9 [PHQ-9]). After model training, the relative influence of individual predictors on depression scores was stratified and analyzed. Model prediction performance was determined by prediction error metrics. RESULTS: The test set mean absolute error was 3.03 (95% confidence interval: 2.91 to 3.14) and 2.55 (95% confidence interval: 2.48 to 2.62) on datasets with audiology-only predictors and all predictors, respectively, on the PHQ-9's 27-point scale. Participants' self-reported frustration when talking to members of family or friends due to hearing loss was the fifth-most influential of all predictors. Of the top 10 most influential audiometric predictors, five were related to social contexts, two for significant noise exposure, two objective audiometric parameters, and one presence of bothersome tinnitus. CONCLUSIONS: Machine learning algorithms can accurately predict PHQ-9 depression scale scores from National Health and Nutrition Examination Survey data. The most influential audiometric predictors of higher scores on a validated depression scale were social dynamics of hearing loss and not objective audiometric testing. Such models could be useful in predicting depression scale scores at the point-of-care in conjunction with a standard audiologic assessment.

A preliminary assessment of guideline adherence and clinical variation in oral cancer treatment: a MarketScan database study.

BACKGROUND: To assess variations in adherence to guideline-recommended processes of care for oral cavity cancer patients. METHODS: Retrospective study using a U.S. healthcare research database (MarketScan). Index diagnoses were considered from 2010 to 2012 with follow-up from 2013 to 2014. Diagnostic and procedure codes were utilized to identify oral cavity patients with a defined treatment modality. Compliance with guideline-recommended processes of care, which included pre-treatment imaging, thyroid-function testing (TFTs), multidisciplinary consultation and gastrostomy-tube insertion rates, were assessed. RESULTS: A total of 2752 patients were identified. Surgery alone was the most common treatment (60.8%), followed by surgery with adjuvant chemoradiotherapy (20.4%) and surgery with adjuvant radiotherapy (18.8%). Head/neck and chest imaging were obtained in 60% and 62.5% of patients respectively. Significant geographical differences in head and neck imaging were observed between North-central (64%), South (58.4%) and West (56.1%) regions (p = 0.026). Differences in chest imaging were also present between North-east (65%) and West (56.8%; p = 0.007). TFTs were obtained in 54.4% of the patients after radiation treatment, and 18.6% of patients had multidisciplinary consultation during the 6 months before and 3 months after initiation of treatment. During the year after treatment initiation, 21.2% of patients underwent G-tube placement, with significantly higher rates in patients receiving triple modality treatment (58%) when compared to surgery plus radiation (27%) and surgery alone (15%; p < 0.01). CONCLUSION: Adherence to evidence-based practices was low based on the database coding. These data suggest a potential to improve adherence and increase the routine use of practices delineated in national clinical practice guidelines. CLINICAL RELEVANCE: This study reflects a suboptimal adherence to guidelines based on the database employed. This study should be considered by healthcare providers and efforts should be maximized to follow the processes of care which have proven to impact on patient's outcomes.

Using Google Trends to Predict Pediatric Respiratory Syncytial Virus Encounters at a Major Health Care System.

To assess whether Google search activity predicts lead-time for pediatric respiratory syncytial virus (RSV) encounters within a major health care system. Internet user search and health system encounter database analysis. Pediatric RSV encounter volumes across all clinics and hospitals in the Duke Health system were tabulated from 2005 to 2016. North Carolina Google user search activity for RSV were obtained over the same time period. Time series analysis was used to compare RSV encounters and search activity. Cross-correlation was used to determine the 'lag' time difference between Google user search interest for RSV and observed Pediatric RSV encounter volumes. Google search activity and Pediatric RSV encounter volumes demonstrated strong seasonality with predilection for winter months. Granger Causality testing revealed that North Carolina RSV Google search activity can predict pediatric RSV encounters at our health system (F = 5.72, p < 0.0001). Using cross-correlation, increases in Google search activity provided lead time of 0.21 weeks (1.47 days) prior to observed increases in Pediatric RSV encounter volumes at our health system. RSV is a common cause of upper airway obstruction in pediatric patients for which pediatric otolaryngologists are consulted. We demonstrate that Google search activity can predict RSV patient interactions with a major health system with a measurable lead-time. The ability to predict when illnesses in a population result in increased health care utilization would be an asset to health system providers, planners and administrators. Prediction of RSV would allow specific care pathways to be developed and resource needs to be anticipated before actual presentation.

AutoAudio: Deep Learning for Automatic Audiogram Interpretation.

Hearing loss is the leading human sensory system loss, and one of the leading causes for years lived with disability with significant effects on quality of life, social isolation, and overall health. Coupled with a forecast of increased hearing loss burden worldwide, national and international health organizations have urgently recommended that access to hearing evaluation be expanded to meet demand. The objective of this study was to develop 'AutoAudio' - a novel deep learning proof-of-concept model that accurately and quickly interprets diagnostic audiograms. Adult audiogram reports representing normal, conductive, mixed and sensorineural morphologies were used to train different neural network architectures. Image augmentation techniques were used to increase the training image set size. Classification accuracy on a separate test set was used to assess model performance. The architecture with the highest out-of-training set accuracy was ResNet-101 at 97.5%. Neural network training time varied between 2 to 7 h depending on the depth of the neural network architecture. Each neural network architecture produced misclassifications that arose from failures of the model to correctly label the audiogram with the appropriate hearing loss type. The most commonly misclassified hearing loss type were mixed losses. Re-engineering the process of hearing testing with a machine learning innovation may help enhance access to the growing worldwide population that is expected to require audiologist services. Our results suggest that deep learning may be a transformative technology that enables automatic and accurate audiogram interpretation.

Vocal Fold Paralysis/Paresis as a Marker for Poor Swallowing Outcomes After Thoracic Surgery Procedures.

(1) To examine the association between vocal fold paresis/paralysis (VFP) and poor swallowing outcomes in a thoracic surgery cohort at the population level, and (2) to assess utilization of ENT/speech-language pathology intervention in these cases. The National Inpatient Sample (NIS) represents a 20% stratified sample of discharges from US hospitals. Using ICD-9 codes, discharges undergoing general thoracic surgical procedures between 2008 and 2013 were identified in the NIS. Sub-cohorts of discharges with VFP and those who utilized ENT/SLP services were also identified. Weighted logistic regression models were used to compare binary outcomes such as dysphagia, aspiration pneumonia, and other complications; generalized linear models with generalized estimating equations (GEE) were used to compare total hospital costs and length of stay (LOS). We identified a weighted estimate of 673,940 discharges following general thoracic surgery procedures. The weighted frequency of VFP was 3738 (0.55%). Compared to those without VFP, patients who discharged with VFP had increased odds of dysphagia (6.56, 95% CI 5.07-8.47), aspiration pneumonia (2.54, 95% CI 1.74-3.70), post-operative tracheotomy (3.10, 95% CI 2.16-4.45), and gastrostomy tube requirement (2.46, 95% CI 1.66-3.64). Discharges with VFP also had a longer length of stay and total hospital costs. Of the discharges with VFP, 15.7% received ENT/SLP intervention. VFP after general thoracic procedures is associated with negative swallowing-related health outcomes and higher costs. Despite these negative impacts, most patients with VFP do not receive ENT/SLP intervention, identifying a potential opportunity for improving adverse swallowing-related outcomes.

Payer database and geospatial analysis to evaluate practice patterns in treating allergy in North Carolina.

OBJECTIVE: The objective of this study was to characterize the delivery of allergy care in North Carolina using a large payer charge database and visualization techniques. STUDY DESIGN: Geospatial database analysis. SETTING: North Carolina State claims database. SUBJECTS & METHODS: Medical data from the 2013 FAIR Health National Private Insurance Claims (FH NPIC) database for North Carolina was mined for CPT codes and charges for allergy testing, and for the preparation and provision of allergen immunotherapy. Provider and patient variables were analyzed. Analyses were performed to compare differences in allergy care delivery. A visualization strategy complemented the analytic approach. RESULTS: 162,037 CPT charge entries were analyzed. Allergy-immunology specialists were the most common provider specialty to perform allergy immunotherapy treatments (68.9%, p<0.05). Among other specialties, there were no significant differences between specialists performing immunotherapy when comparing otolaryngology, family practice, and internal medicine (16.3%; 4.6%; 2.6%; p>0.05). Providers with an M.D. degree were the most common provider type. The three most commonly treated diagnoses were allergic rhinitis variants. Females were more likely to receive allergy treatments versus males (55.9% vs. 51.5%; p<0.001), and were more likely to receive allergy testing (65.3% vs. 34.7%: p<0.005). Internal medicine providers charged higher than any other specialist type (p<0.05) for allergy immunotherapy. CONCLUSIONS: Using a large payer database coupled with visualization techniques was an efficient approach to characterizing the state-wide provision patterns of allergy diagnostic and therapy services in North Carolina. This first tier approach to efficiently exploring questions and describing populations is valuable.

Cost-effectiveness analysis of a non-contrast screening MRI protocol for vestibular schwannoma in patients with asymmetric sensorineural hearing loss.

PURPOSE: We aimed to determine if a non-contrast screening MRI is cost-effective compared to a full MRI protocol with contrast for the evaluation of vestibular schwannomas. METHODS: A decision tree was constructed to evaluate full MRI and screening MRI strategies for patients with asymmetric sensorineural hearing loss. If a patient were to have a positive screening MRI, s/he received a full MRI. Vestibular schwannoma prevalence, MRI specificity and sensitivity, and gadolinium anaphylaxis incidence were obtained through literature review. Institutional charge data were obtained using representative patient cohorts. One-way and probabilistic sensitivity analyses were completed to determine CE model threshold points for MRI performance characteristics and charges. RESULTS: The mean charge for a full MRI with contrast was significantly higher than a screening MRI ($4089 ± 1086 versus $2872 ± 741; p < 0.05). The screening MRI protocol was more cost-effective than a full MRI protocol with a willingness-to-pay from $0 to 20,000 USD. Sensitivity analyses determined that the screening protocol dominated when the screening MRI charge was less than $4678, and the imaging specificity exceeded 78.2%. The screening MRI protocol also dominated when vestibular schwannoma prevalence was varied between 0 and 1000 in 10,000 people. CONCLUSION: A screening MRI protocol is more cost-effective than a full MRI with contrast in the diagnostic evaluation of a vestibular schwannoma. A screening MRI likely also confers benefits of shorter exam time and no contrast use. Further investigation is needed to confirm the relative performance of screening protocols for vestibular schwannomas.

Quality of Life and Cost-Effectiveness of Cochlear Implants: A Narrative Review.

OBJECTIVES: To review evidence regarding the health-related quality of life (HRQoL) and cost-effectiveness of unilateral and bilateral cochlear implantation (CI) among children and adults with severe-to-profound hearing loss. STUDY DESIGN: Narrative review. METHODS: Publications related to quality of life (QoL) and costs of care in CI were acquired through searches in English-language databases. Studies were included if they had identified the HRQoL attainment, cost of care, cost-utility, or cost-effectiveness associated with CI. RESULTS: 57 studies were critically reviewed. The QoL outcome metrics used in these articles were divided into 2 categories - generic and condition specific. In studies investigating children, many reported no significant difference in QoL attainment between CI recipients and normal-hearing peers. In adults, significant improvements in QoL after implantation and higher QoL than in their nonimplanted (hearing-aided) peers were frequently reported. Studies involving an older adult cohort reported significant improvement in QoL after implantation, which was often independent of audiological performance. Overall, the calculated cost-utility ratios consistently met the threshold of cost acceptance, indicating acceptable values for expenditures on CI. CONCLUSIONS: Considerable work has been done on the QoL attainment and health economic implications of CI. Unilateral CI across all age groups leads to reported sustained benefits in the recipients' overall and disease-specific QoL. With increased cost associated with bilateral CI, further study is needed to characterize its costs and benefits with respect to the recipients' health, well-being, and contributions to society.

Mini Review of the Cost-Effectiveness of Unilateral Osseointegrated Implants in Adults: Possibly Cost-Effective for the Correct Indication.

An osseointegrated implant (e.g. bone-anchored hearing aid, BAHA) is a surgically implantable device for unilateral sensorineural and unilateral or bilateral conductive hearing loss in patients who otherwise cannot use or do not prefer a conventional air conduction hearing aid (ACHA). The specific indications for an osseointegrated implant are evolving and dependent upon the country or regulatory body overseeing the provision of these devices. However, there are general groups of patients who would be likely to benefit, one such group being patients with congenital aural atresia. Given the anatomical aberrancies with aural atresia, these subjects cannot wear ACHAs. Another group of patients who may benefit from an osseointegrated implant over an ACHA are patients with chronically draining otological infections. As the provision of an osseointegrated implant requires a surgical procedure, there are inherent direct and indirect costs associated with its use beyond those required for an ACHA. Consideration of outcomes and cost-effectiveness for the osseointegrated implant versus the ACHA is prudent prior to making policy decisions in a setting of limited health care resources. We performed a mini review on all available cost-effectiveness analyses of osseointegrated implants published in Medline. There are only 2 contemporary cost-effectiveness analyses published to date. There is limited quality of life data available for patients living with an osseointegrated implant. As a result, the cost-effectiveness of the osseointegrated implant, specifically the BAHA, compared to conventional hearing aid devices remains unclear. However, there are clear indications for the BAHA when a standard hearing aid cannot be used (e.g. chronic draining ear) or in single-sided severe-to-profound hearing loss with reasonable hearing in the contralateral ear. The BAHA should not be considered interchangeable with the ACHA with regard to cost-effectiveness, but rather considered as an effective option for the patient for the correct indication.

Preoperative Lumbar Drain Use During Acoustic Neuroma Surgery and Effect on CSF Leak Incidence.

OBJECTIVE: To determine if preoperative lumbar drain (LD) use reduces the incidence of postoperative cerebrospinal fluid (CSF) leak in patients undergoing acoustic neuroma resection. METHODS: Retrospective review of 282 patients presenting for acoustic neuroma resection between 2005 and 2014. RESULTS: Two hundred and eighty-two patients had a mean tumor size of 19.1 mm ± 10.2 mm. Twenty-nine (10.3%) patients developed a postoperative CSF leak. Two hundred and twenty patients (78.0%) received a preoperative LD, and 20 (9.1%) developed a CSF leak. Sixty-two (22.0%) patients did not receive a preoperative LD, and 9 (14.5%) developed a CSF leak. No significant difference in CSF leak frequency was observed with use versus no use of a LD (P = .23). Fifteen (5.3%) patients with an LD placed had a complication related to the LD. No significant difference in CSF leak frequency was observed with patient age, neurofibromatosis type-2 diagnosis, tumor size, or sidedness. CONCLUSIONS: Postoperative CSF leaks are among the most common complications of acoustic neuroma microsurgery. No formal guidelines exist for elective placement of a preoperative LD to lower the incidence of CSF leaks. Our reported CSF leak incidence with preoperative LD placement is not significantly lower than without LD use, and there is a complication rate associated with LD use.

Rare cases of large parapharyngeal lipomas.

Primary tumors of the parapharyngeal space are extremely rare, and lipomas are among the least common primary parapharyngeal space masses. Parapharyngeal lipomas typically present as a painless neck mass, and some may present with neurologic deficits or vascular compromise attributed to the lipomas' mass effect on nearby neurovascular structures. We report long term follow-up of two large parapharyngeal lipomas. One lesion was managed expectantly, and the other was managed with a partial transcervical excision. We demonstrate that conservative management and long term patient follow-up may be reasonable if the patient is asymptomatic and liposarcoma is ruled out. Considering the uncertainty in the need for removal, the management strategy for the individual patient is best to be tailored to their clinical presentation.

Utility of Daily Mobile Tablet Use for Residents on an Otolaryngology Head & Neck Surgery Inpatient Service.

The objective of this study was to investigate the utility of electronic tablets and their capacity to increase hospital floor productivity, efficiency, improve patient care information safety, and to enhance resident education and resource utilization on a busy Otolaryngology - Head & Neck Surgery inpatient service. This was a prospective cohort study with a 2-week pre-implementation period with standard paper census lists without mobile tablet use, and a 2-week post-implementation period followed with electronic tablets used to place orders, look up pertinent clinical data, educate patients as appropriate, and to record daily to-dos that would previously be recorded on paper. The setting for the study was Duke University Medical Center in Durham, North Carolina, with 13 Otolaryngology residents comprising the study population. The time for inpatient rounding was shorter with the use tablets (p = 0.037). There was a non-significant trend in the number of times a resident had to leave rounds to look up a clinical query on a computer, with less instances occurring in the post-implementation study period. The residents felt that having a tablet facilitated more detailed and faster transfer of information, and improved ease of documentation in the medical record. Seventy percent felt tablets helped them spend more time with patients, 70 % could spend more time directly involved in rounds because they could use the tablet to query information at point-of-care, and 80 % felt tablets improved morale. The utility of a mobile tablet device coupled with the electronic health record appeared to have both quantitative and qualitative improvements in efficiency, increased time with patients and attendance at academic conferences. Tablets should be encouraged but not mandated for clinical and educational use.

Access to and Uptake of Cochlear Implantation Among Children in North Carolina.

BACKGROUND: Cochlear implantation (CI) is a highly effective intervention for children with advanced hearing loss who cannot benefit from amplification. Despite the established benefits of CI, it is likely that not all children who are potential candidates for CI receive this intervention. The purpose of this study was to determine the percentage of North Carolina children who are candidates for and end up undergoing CI, and to detect whether barriers exist that prevent access to care for unimplanted candidates. METHODS: This study was a retrospective analysis of 1,501 children whose families were served by BEGINNINGS from January 1, 2009 through December 31, 2013. All families of children identified as potential CI candidates who were able to participate in the study (n = 141) were contacted by BEGINNINGS parent educators who queried parents about their child's use of technology and any reasons for lack of use of technology. RESULTS: Overall, 60.9% of children diagnosed with profound, severe-profound, severe, moderate-severe, or moderate-profound hearing loss received at least 1 cochlear implant. For children with profound hearing loss, 88.9% had a least 1 cochlear implant. Common reasons for the decision not to perform CI included parental preference and anatomical issues unfavorable to CI. LIMITATIONS: Some information was not included in the database, including socioeconomic status and the child's age at the time of intervention. CONCLUSION: The rate of CI for North Carolina children who have advanced hearing loss is greater than 60% and significantly higher for children with greater degrees of impairment. No significant financial or geographic barriers to CI were identified. We hypothesize that the high rate of CI for appropriate candidates in North Carolina is due in part to parental access to counseling and education provided through BEGINNINGS.

Provider and patient drivers of ototopical antibiotic prescription variability.

OBJECTIVE: To determine if providers prescribe more affordable topical antibacterial therapy for patients who are economically disadvantaged or come from economically disadvantaged communities. STUDY DESIGN: Prescription drug database review. SETTING: Large academic hospital network. SUBJECTS AND METHODS: Ototopical prescription records of 2416 adults and children presenting with acute and chronic otologic infections from 2009 to 2013 were reviewed. Prescription, patient, provider, and institution variables including diagnosis, prescription type, demographics, health insurance status, healthcare provider type and setting were analyzed. RESULTS: Otitis externa and acute otitis media were the most common diagnoses. Non-OHNS (Otolaryngology-Head and Neck Surgery) providers served 82% of all patients. OHNS providers prescribed proportionally less fluoroquinolone, and more brand-name antibiotics compared to non-OHNS providers. Adults were more likely to receive a non-fluoroquinolone antibiotic and a generic prescription versus pediatric patients. Patients who self-identified as 'white' ethnicity received proportionally more fluoroquinolone prescriptions than patients who identified as 'non-white,' but there was no difference in provider type. The proportion of fluoroquinolone prescriptions was significantly higher in patients from low-poverty counties, however poverty level was not associated with patients seeing a particular provider type. The majority of our patients had commercial insurance, followed by Medicaid. Medicare patients had the lowest proportion of fluoroquinolone antibiotic prescriptions, and were less likely to receive fluoroquinolone prescriptions versus commercial insurance. Non-insured patients received proportionally more generic versus brand prescriptions than insured patients. CONCLUSION: Our results indicate potential provider, patient demographic, and financial factors producing considerable variability in the prescribing patterns for topical antibiotics for common otologic infections.

The Integration and Impact of Artificial Intelligence in Otolaryngology-Head and Neck Surgery: Navigating the Last Mile.

Incorporating artificial Intelligence and machine learning into otolaryngology requires careful data handling, security, and ethical considerations. Success depends on interdisciplinary cooperation, consistent innovation, and regulatory compliance to improve clinical outcomes, provider experience, and operational effectiveness.

A Proof-of-Concept Computer Vision Approach for Measurement of Tympanic Membrane Perforations.

OBJECTIVE: Size, an important characteristic of a tympanic membrane perforation (TMP), is commonly assessed with gross estimation via visual inspection, a practice which is prone to inaccuracy. Herein, we demonstrate feasibility of a proof-of-concept computer vision model for estimating TMP size in a small set of perforations. METHODS: An open-source deep learning architecture was used to train a model to segment and calculate the area of a perforation and the visualized tympanic membrane (TM) in a set of endoscopic images of mostly anterior and relatively small TMPs. The model then computed relative TMP size by calculating the ratio of perforation area to TM area. Model performance on the test dataset was compared to ground-truth manual annotations. In a validation survey, otolaryngologists were tasked with estimating the size of TMPs from the test dataset. The primary outcome was the average absolute error of model size predictions and clinician estimates compared to sizes determined by ground-truth manual annotations. RESULTS: The model's average absolute error for size predictions was a 0.8% overestimation for all test perforations. Conversely, among the 38 survey respondents, the average clinician error was a 11.0% overestimation (95% CI, 5.2-16.7%, p = 0.003). CONCLUSIONS: In a small sample of TMPs, we demonstrated a computer vision approach for estimating TMP size is feasible. Further validation studies must be done with significantly larger and more heterogenous datasets. LEVEL OF EVIDENCE: N/A Laryngoscope, 134:2906-2911, 2024.

Proof-of-Concept Computer Vision Model for Instrument and Anatomy Detection During Transcanal Endoscopic Ear Surgery.

High-definition video captured during transcanal endoscopic ear surgery (TEES) can serve as imaging data for computer vision algorithms. This report describes a proof-of-concept model for automated anatomy and instrument detection during TEES.