BACKGROUND Deep learning (DL)–based artificial intelligence may have different diagnostic characteristics than human experts in medical diagnosis. As a data-driven knowledge system, heterogeneous population incidence in the clinical world is considered to cause more bias to DL than clinicians. Conversely, by experiencing limited numbers of cases, human experts may exhibit large interindividual variability. Thus, understanding how the 2 groups classify given data differently is an essential step for the cooperative usage of DL in clinical application. OBJECTIVE This study aimed to evaluate and compare the differential effects of clinical experience in otoendoscopic image diagnosis in both computers and physicians exemplified by the class imbalance problem and guide clinicians when utilizing decision support systems. METHODS We used digital otoendoscopic images of patients who visited the outpatient clinic in the Department of Otorhinolaryngology at Severance Hospital, Seoul, South Korea, from January 2013 to June 2019, for a total of 22,707 otoendoscopic images. We excluded similar images, and 7500 otoendoscopic images were selected for labeling. We built a DL-based image classification model to classify the given image into 6 disease categories. Two test sets of 300 images were populated: balanced and imbalanced test sets. We included 14 clinicians (otolaryngologists and nonotolaryngology specialists including general practitioners) and 13 DL-based models. We used accuracy (overall and per-class) and kappa statistics to compare the results of individual physicians and the ML models. RESULTS Our ML models had consistently high accuracies (balanced test set: mean 77.14%, SD 1.83%; imbalanced test set: mean 82.03%, SD 3.06%), equivalent to those of otolaryngologists (balanced: mean 71.17%, SD 3.37%; imbalanced: mean 72.84%, SD 6.41%) and far better than those of nonotolaryngologists (balanced: mean 45.63%, SD 7.89%; imbalanced: mean 44.08%, SD 15.83%). However, ML models suffered from class imbalance problems (balanced test set: mean 77.14%, SD 1.83%; imbalanced test set: mean 82.03%, SD 3.06%). This was mitigated by data augmentation, particularly for low incidence classes, but rare disease classes still had low per-class accuracies. Human physicians, despite being less affected by prevalence, showed high interphysician variability (ML models: kappa=0.83, SD 0.02; otolaryngologists: kappa=0.60, SD 0.07). CONCLUSIONS Even though ML models deliver excellent performance in classifying ear disease, physicians and ML models have their own strengths. ML models have consistent and high accuracy while considering only the given image and show bias toward prevalence, whereas human physicians have varying performance but do not show bias toward prevalence and may also consider extra information that is not images. To deliver the best patient care in the shortage of otolaryngologists, our ML model can serve a cooperative role for clinicians with diverse expertise, as long as it is kept in mind that models consider only images and could be biased toward prevalent diseases even after data augmentation.

Background and Objectives Previous studies reported abnormalities in MRI as a poor prognostic indicator of sudden sensorineural hearing loss (SSNHL). Since abnormalities in three-dimensional (3D) fluid-attenuated inversion recovery (FLAIR) are strongly correlated with the initial hearing function, the prognostic value of the 3D FLAIR images should be carefully evaluated to avoid collinearity. We aimed to evaluate abnormalities on the 3D FLAIR images as an independent prognostic factor in the matched SSNHL groups. Subjects and MethodWe retrospectively reviewed medical records of 179 patients with SSNHL who underwent temporal MRI, including the 3D FLAIR sequence, between January 2015 and December 2019. Patients were divided based on the presence of cochlear abnormalities on the 3D FLAIR images. Hearing prognosis was evaluated with and without matching for initial hearing and treatment interval. Results The groups were similar in sex (p=0.091), age (p=0.925), treatment interval (p= 0.216), and MRI interval (p=0.828). Notably, patients with cochlear abnormalities on the 3D FLAIR images showed distinctly more severe hearing loss (p<0.001) at the initial pure tone average (PTA) assessment and poorer outcomes (p<0.001) compared to those without abnormality. After matching for initial hearing and treatment interval, the hearing outcome, measured by PTA, was similar between the groups (p=0.681). Conclusion Cochlear signal abnormality in 3D FLAIR MRI was associated with poor initial hearing. However, it did not affect hearing recovery outcomes when the groups were matched.

Background and Objectives: Osteomas are the most common benign tumors of the nasal cavity and paranasal sinuses (PNSs). In this study, clinical features and imaging findings were analyzed in patients with osteoma confirmed by ostiomeatal unit (OMU) computed tomography (CT) and PNS CT, and the surgical treatment performed at our hospital was introduced.Methods: The Severance Clinical Research Analysis Portal (SCRAP) service of Severance Hospital was used to collect research data. A total of 128 cases of osteomas of the nasal cavity or PNSs confirmed by OMU CT or PNS CT was retrospectively reviewed, including the location and size of the osteoma, clinical features, accompanying findings on imaging tests, and cases of surgical treatment.Results: In this study, osteomas were found in about 0.55% of patients who underwent computed tomography. Osteomas were most frequently found in the ethmoid sinus, followed by the frontal sinus, fronto-ethmoid sinus, maxillary sinus, intranasal sphenoid sinus, and maxillary sinus-ethmoid sinus. Patients with osteomas complained of symptoms such as rhinorrhea, postnasal drip, nasal congestion, hyposmia, headache, visual disturbance, and lacrimal duct obstruction.Conclusion: Surgical treatment was considered for patients presenting with severe headache, visual field symptoms, or accompanying rhinosinusitis. Surgery was performed by endoscopic or external approaches depending on location and size of the osteoma.

Interpreting the relationship between different taste function tests of different stimuli, such as chemical and electrical stimulation, is still poorly understood. This study aims to analyze visually as well as quantitatively how to interpret the relationship of results between taste function tests using different stimuli. Patients who underwent the whole mouth test and Electrogustometry (EGM) at a tertiary medical center between August 2018 and December 2018 were reviewed retrospectively with electronic medical records. Of the 110 patients, a total of 86 adults who self-reported that their taste function was normal through a questionnaire were enrolled. EGM measured the thresholds of the chorda tympani (CT) and glossopharyngeal nerve (GL) area of the tongue. The whole mouth test measured detection and recognition thresholds for sweet, salty, bitter, sour, and umami taste. Statistical analyses of Pearson’s, Spearman’s rank and polyserial correlation and multidimensional scaling (MDS) was performed. The EGM threshold for the average value of both CT regions and the recognition threshold of the whole mouth test were significantly correlated in sweet, salty, bitter, and sour taste (r = 0.244–0.398, P < 0.05), and the detection threshold was correlated only significant in sweet (r = 0.360, P = 0.007). In the MDS analysis results, the three-dimensional (D) solution was chosen over the 2-D solution because of the lower stress. Detection-, recognition threshold of whole mouth test and EGM thresholds of CT and GL area, those were standardized by Z-score, formed well-distinguished sections in the MDS analyses. The EGM threshold of the CT area was closer to the detection and recognition thresholds than the EGM threshold of the GL area. In general, the EGM threshold was closer to the recognition threshold than the detection threshold for each taste. Overall, visualization of the relationship of whole mouth test and EGM by MDS was in good agreement with quantitative analysis. EGM and whole mouth test seem to reflect different aspects of taste. However, when interpreting the EGM results, the EGM threshold of the CT area will show more similarity to the recognition threshold than the detection threshold for the whole mouth test.

Background and Objectives: Some reports propose an increased risk of otitis media and hearing impairment after total laryngectomy. However, the incidence of otitis media following laryngectomy and the mechanism remain unclear. This study aimed to identify the incidence and risk factors of otitis media after total laryngectomy.Subjects and Methods: This retrospective cohort study assessed 77 patients who underwent total laryngectomy from 2010 to 2020 in a tertiary referral center. Serial imaging studies (computed tomography [CT], magnetic resonance imaging, and positron emission tomography-CT) were used to assess otitis media.Results: The study enrolled 58 patients (mean age, 67.0±7.7 years; male, 56 [96.6%]); nine (15.5%) underwent a gastrostomy tube (four preoperatively and five postoperatively). Otitis media was confirmed in seven (12.1%) patients. Gastrostomy tube insertion was the only significant risk factor for otitis media (p=0.012). Of the nine patients who underwent gastrostomy tube insertion, four developed otitis media; all four had the procedure after laryngectomy.Conclusions: This study found an increased incidence of otitis media after total laryngectomy. Swallowing difficulties likely contribute to otitis media as it occurred more frequently in patients requiring postoperative gastrostomy tube insertion.