The present study aimed to develop a machine learning network to diagnose middle ear diseases with tympanic membrane images and to identify its assistive role in the diagnostic process. The medical records of subjects who underwent ear endoscopy tests were reviewed. From these records, 2272 diagnostic tympanic membranes images were appropriately labeled as normal, otitis media with effusion (OME), chronic otitis media (COM), or cholesteatoma and were used for training. We developed the “ResNet18 + Shuffle” network and validated the model performance. Seventy-one representative cases were selected to test the final accuracy of the network and resident physicians. We asked 10 resident physicians to make diagnoses from tympanic membrane images with and without the help of the machine learning network, and the change of the diagnostic performance of resident physicians with the aid of the answers from the machine learning network was assessed. The devised network showed a highest accuracy of 97.18%. A five-fold validation showed that the network successfully diagnosed ear diseases with an accuracy greater than 93%. All resident physicians were able to diagnose middle ear diseases more accurately with the help of the machine learning network. The increase in diagnostic accuracy was up to 18% (1.4% to 18.4%). The machine learning network successfully classified middle ear diseases and was assistive to clinicians in the interpretation of tympanic membrane images.

Previous studies have demonstrated the potential for using smartwatches with a built-in accelerometer as feedback devices for high-quality chest compression during cardiopulmonary resuscitation. However, to the best of our knowledge, no previous study has reported the effects of this feedback on chest compressions in action. A randomized, parallel controlled study of 40 senior medical students was conducted to examine the effect of chest compression feedback via a smartwatch during cardiopulmonary resuscitation of manikins. A feedback application was developed for the smartwatch, in which visual feedback was provided for chest compression depth and rate. Vibrations from smartwatch were used to indicate the chest compression rate. The participants were randomly allocated to the intervention and control groups, and they performed chest compressions on manikins for 2 min continuously with or without feedback, respectively. The proportion of accurate chest compression depth (≥5 cm and ≤6 cm) was assessed as the primary outcome, and the chest compression depth, chest compression rate, and the proportion of complete chest decompression (≤1 cm of residual leaning) were recorded as secondary outcomes. The proportion of accurate chest compression depth in the intervention group was significantly higher than that in the control group (64.6±7.8% versus 43.1±28.3%; p = 0.02). The mean compression depth and rate and the proportion of complete chest decompressions did not differ significantly between the two groups (all p>0.05). Cardiopulmonary resuscitation-related feedback via a smartwatch could provide assistance with respect to the ideal range of chest compression depth, and this can easily be applied to patients with out-of-hospital arrest by rescuers who wear smartwatches.

Healthcare providers in emergency departments should wear respirators for infection protection. However, the wearer's vigorous movements during cardiopulmonary resuscitation may affect the protective performance of the respirator. Herein, we aimed to assess the effects of chest compressions (CCs) on the protective performance of respirators. This crossover study evaluated 30 healthcare providers from 1 emergency department who performed CC with real-time feedback. The first, second, and third groups started with a cup-type, fold-type, and valve-type respirator, respectively, after which the respirators were randomized for each group. The fit factors were measured using a quantitative fit testing device before and during the CC in each experiment. The protection rate was defined as the proportion of respirators achieving a fit factor ≥100. The fold-type respirator had a significantly greater protection rate at baseline (100.0% ± 0.0%) compared to the cup-type (73.6% ± 39.6%, P = .003) and valve-type respirators (87.5% ± 30.3%, P = .012). During the CC, the fit factor values significantly decreased for the cup-type (44.9% ± 42.8%, P < .001) and valve-type respirators (59.5% ± 41.7%, P = .002), but not for the fold-type respirator (93.2% ± 21.7%, P = .095). The protective performances of respirators may be influenced by CC. Healthcare providers should identify the respirator that provides the best fit for their intended tasks.