Predictive factors of acute sensorineural hearing loss in adult Japanese patients for clinical application by primary care doctors: a cross-sectional study.

BACKGROUND: Several methods are used for hearing loss screening; however, their benefits are uncertain. In this study, we aimed to determine the predictive factors of acute sensorineural hearing loss for clinical application by primary care doctors. METHODS: This retrospective, cross-sectional study included 365 patients with acute sensorineural hearing loss without prior therapy. The patients' clinical data, demographic information, and medical histories were obtained, and they were asked about comorbidities. In addition, we assessed lifestyle factors such as stress level, alcohol consumption, marital status, and socioeconomic level. Logistic regression analysis was performed to investigate the diagnostic predictive ability of the selected factors associated with acute sensorineural hearing loss. The hearing levels of all patients were evaluated using pure tone audiometry. RESULTS: We identified significant predictive factors for acute sensorineural hearing loss. The absence of hyperacusis was a predictive factor for sudden sensorineural hearing loss. Younger age, female sex, and marital status were predictive factors for acute low-tone hearing loss. High body mass index, high socioeconomic level, low alcohol consumption, high stress level, hyperacusis, and vertigo/dizziness were predictive factors for Ménière's disease. High body mass index and ear fullness were predictive factors for perilymph fistula. Low stress level was a predictive factor for acoustic tumours. CONCLUSIONS: Our findings can be used to distinguish between the types of acute sensorineural hearing loss. Symptoms, physical status, and lifestyle factors identified during this study are useful markers for predicting acute sensorineural hearing loss occurrence.

Application of artificial intelligence using a convolutional neural network for detecting cholesteatoma in endoscopic enhanced images.

OBJECTIVE: We examined whether artificial intelligence (AI) used with the novel digital image enhancement system modalities (CLARA+CHROMA, SPECTRA A, and SPECTRA B) could distinguish the cholesteatoma matrix, cholesteatoma debris, and normal middle ear mucosa, and observe the middle ear cavity during middle ear cholesteatoma surgery. METHODS: A convolutional neural network (CNN) was trained with a set of images chosen by an otologist. To evaluate the diagnostic accuracy of the constructed CNN, an independent test data set of middle ear images was collected from 14 consecutive patients with 26 cholesteatoma matrix lesions, who underwent transcanal endoscopic ear surgery at a single hospital from August 2018 to September 2019. The final test data set included 58 total images, with 1‒5 images from each modality for each case. RESULTS: The CNN required only 10 s to analyze more than 58 test images. Using SPECTRA A and SPECTRA B, the CNN correctly diagnosed 15 and 15 of 26 cholesteatoma matrix lesions, with a sensitivity of 34.6% and 42.3%, and with a specificity of 81.3% and 87.5%, respectively. CONCLUSION: Our preliminary study revealed that AI and novel imaging modalities are potentially useful tools for identifying and visualizing the cholesteatoma matrix during endoscopic ear surgery. The diagnostic ability of the CNN is not yet appropriate for implementation in daily clinical practice, based on our study findings. However, in the future, these techniques and AI tools could help to reduce the burden on surgeons and will facilitate telemedicine in remote and rural areas, as well as in developing countries where the number of surgeons is limited.