Search

Learning-based data structures, such as a learned Bloom filter and a learned functional Bloom filter (L-FBF), have recently been proposed to replace traditional structures. However, using these structures for dynamic data processing is difficult because a specific element cannot be deleted from a trained model. A counting Bloom filter with return values (rCBF) is a more efficient key–value structure than a functional Bloom filter (FBF) for repetitive insertions and deletions. In this study, we propose a learned rCBF (L-rCBF) comprising a model, a Bloom filter, and an rCBF and the deletion algorithm for the L-rCBF. To delete a specific element from the L-rCBF, two different operations are performed according to four different cases. In the experiments, the proposed L-rCBF is compared with a single rCBF and an L-FBF in terms of undeletables and search failures, and this comparison is conducted using two different models. In addition, we present a theoretical analysis of the rCBF with experimental results to demonstrate that a structure with an rCBF is more suitable for dynamic data than a structure with an FBF. [ABSTRACT FROM AUTHOR]

Most patients with stroke suffer from complications and these include dysphagia. Dysphagia can cause malnutrition, and malnutrition affects prognosis and recovery. However, there is a lack of accurate studies on the nutritional status of stroke patients with dysphagia and its associated factors in different phases of stroke. This study retrospectively investigated 620 stroke patients who underwent a videofluoroscopic swallowing study (VFSS) due to dysphagia, from March 2018 to February 2021. The study aimed to evaluate the nutritional state and associated factors of malnutrition in acute and subacute stroke patients with dysphagia. Serum albumin and percentage of current weight to ideal weight were used to determine nutritional status. Malnutrition was observed in 58.9 and 78.9% of acute and subacute stroke patients. Exact logistic regression analysis revealed that old age and high penetration–aspiration scale score were significantly associated factors for malnutrition in patients with acute stroke. Old age, stroke history, bilateral hemiplegia, high modified Rankin score, low Korean Mini-Mental State Examination, pneumonia, and high functional dysphagia score were significantly associated factors for malnutrition in patients with subacute stroke. Patients with these associated factors in each phase of stroke require active nutritional assessment and care to decrease the risk of malnutrition. [ABSTRACT FROM AUTHOR]

A machine learning platform operated without coding knowledge (Teachable machine®) has been introduced. The aims of the present study were to assess the performance of the Teachable machine® for diagnosing tympanic membrane lesions. A total of 3024 tympanic membrane images were used to train and validate the diagnostic performance of the network. Tympanic membrane images were labeled as normal, otitis media with effusion (OME), chronic otitis media (COM), and cholesteatoma. According to the complexity of the categorization, Level I refers to normal versus abnormal tympanic membrane; Level II was defined as normal, OME, or COM + cholesteatoma; and Level III distinguishes between all four pathologies. In addition, eighty representative test images were used to assess the performance. Teachable machine® automatically creates a classification network and presents diagnostic performance when images are uploaded. The mean accuracy of the Teachable machine® for classifying tympanic membranes as normal or abnormal (Level I) was 90.1%. For Level II, the mean accuracy was 89.0% and for Level III it was 86.2%. The overall accuracy of the classification of the 80 representative tympanic membrane images was 78.75%, and the hit rates for normal, OME, COM, and cholesteatoma were 95.0%, 70.0%, 90.0%, and 60.0%, respectively. Teachable machine® could successfully generate the diagnostic network for classifying tympanic membrane. [ABSTRACT FROM AUTHOR]

Background: This study aimed to develop an algorithm for multilabel classification according to the distance from carina to endotracheal tube (ETT) tip (absence, shallow > 70 mm, 30 mm ≤ proper ≤ 70 mm, and deep position < 30 mm) with the application of automatic segmentation of the trachea and the ETT on chest radiographs using deep convolutional neural network (CNN). Methods: This study was a retrospective study using plain chest radiographs. We segmented the trachea and the ETT on images and labeled the classification of the ETT position. We proposed models for the classification of the ETT position using EfficientNet B0 with the application of automatic segmentation using Mask R-CNN and ResNet50. Primary outcomes were favorable performance for automatic segmentation and four-label classification through five-fold validation with segmented images and a test with non-segmented images. Results: Of 1985 images, 596 images were manually segmented and consisted of 298 absence, 97 shallow, 100 proper, and 101 deep images according to the ETT position. In five-fold validations with segmented images, Dice coefficients [mean (SD)] between segmented and predicted masks were 0.841 (0.063) for the trachea and 0.893 (0.078) for the ETT, and the accuracy for four-label classification was 0.945 (0.017). In the test for classification with 1389 non-segmented images, overall values were 0.922 for accuracy, 0.843 for precision, 0.843 for sensitivity, 0.922 for specificity, and 0.843 for F1-score. Conclusions: Automatic segmentation of the ETT and trachea images and classification of the ETT position using deep CNN with plain chest radiographs could achieve good performance and improve the physician's performance in deciding the appropriateness of ETT depth. [ABSTRACT FROM AUTHOR]

As a challenging attempt to replace a traditional data structure with a learned model, this paper proposes a learned functional Bloom filter (L-FBF) for a key–value storage. The learned model in the proposed L-FBF learns the characteristics and the distribution of given data and classifies each input. It is shown through theoretical analysis that the L-FBF provides a lower search failure rate than a single FBF in the same memory size, while providing the same semantic guarantees. For model training, character-level neural networks are used with pretrained embeddings. In experiments, four types of different character-level neural networks are trained: a single gated recurrent unit (GRU), two GRUs, a single long short-term memory (LSTM), and a single one-dimensional convolutional neural network (1D-CNN). Experimental results prove the validity of theoretical results, and show that the L-FBF reduces the search failures by 82.8% to 83.9% when compared with a single FBF under the same amount of memory used. [ABSTRACT FROM AUTHOR]

BACKGROUND: Obstructive sleep apnea (OSA) could influence peripheral vestibular function adversely via intermittent hypoxia and its consequences. OBJECTIVE: This study aimed to evaluate the risk of peripheral vestibular disorders in OSA using a nationwide population-based retrospective cohort study. METHODS: The National Health Insurance Service–National Sample Cohort represents the entire Korean population from 2002 to 2015. OSA was defined as individuals who had used medical services twice or more under a diagnosis of OSA(G47.33 in ICD-10). A comparison cohort consisted of socio-demographically matched non-OSA subjects in a ratio of 1:4. The incidences of benign paroxysmal positional vertigo(BPPV), Meniere's disease, and vestibular neuritis were evaluated in each cohort. RESULTS: A total of 2,082 individuals with OSA and 8,328 matched non-OSA subjects were identified. The incidence rates(IRs) of peripheral vertigo in OSA and non-OSA were 149.86 and 23.88 per 10,000 persons, respectively (Ratio of IR, IRR = 6.28, 95%CI 4.89 to 8.08). In multivariable analysis, the risk of peripheral vertigo was significantly higher in OSA(adjusted HR = 6.64, 95%CI 5.20 to 8.47), old age(adjusted HR = 1.03, 95%CI 1.02 to 1.04), female sex(adjusted HR = 1.92, 95%CI 1.48 to 2.50), and comorbidities(adjusted HR = 1.09, 95%CI 1.003 to 1.19). The IRRs of each vestibular disorder in the two groups were 7.32(95%CI 4.80 to 11.33) for BPPV, 3.61(95%CI 2.24 to 5.81) for Meniere's disease, and 9.51(95%CI 3.97 to 25.11) for vestibular neuritis. CONCLUSIONS: Subjects diagnosed with OSA had a higher incidence of peripheral vestibular disorders than those without OSA, according to national administrative claims data. It is recommended to take peripheral vertigo into account when counseling OSA. [ABSTRACT FROM AUTHOR]

Objectives: Foreign bodies (FBs) in external auditory canal (EAC) can occur at any age but are especially common in children and adolescents. The aims of the study were to evaluate the clinical characteristics of EAC FBs in children and adolescents and to discuss their proper management. Methods: A retrospective medical chart review was performed on patients who were treated for EAC FBs in a tertiary referral center from January 2007 to December 2017. We investigated the parameters of age, sex, type of FB, sedation method, complications, and removal strategy. Results: Of 284 patients, 176 (62.0%) were male. Peak incidences were noted at the ages of 3 to 6 years and 17 to 18 years. The frequently observed FBs were insects, followed by air-gun pellets, cotton balls, marbles, and earrings. Forty-one (14.4%) EAC FBs were removed with the naked eye, while 243 (85.6%) were removed under microscope in the otolaryngology department. Of these patients, 23 (9.5%) were administered sedatives and 1 (4.1%) needed general anesthesia. During or after the removal procedure, 4 (15.5%) patients had complications of EAC abrasion or laceration (42, 14.8%), and tympanic membrane perforation (2, 0.7%). Conclusions: External auditory canal FB showed a distinct characteristic of incidence regarding age, related to removal strategies. Characteristics of FB must be considered for safe removal. [ABSTRACT FROM AUTHOR]

Background: To compare long-term effects of antigravity treadmill (AGT) combined with conventional rehabilitation (CR) and CR after hip fracture in patients with sarcopenia.Methods: Forty-five patients were randomly allocated to AGT combined with CR (experimental group) or CR (control group) for 10 consecutive working days. Participants were evaluated prior to treatment, 3 weeks, 3 months, and 6 months after treatment. Outcome measurement included Koval walking ability scores functional ambulatory category (FAC), Berg Balance Scale (BBS), Korean version of Mini-Mental State Examination, Euro Quality of Life Questionnaire Five-Dimensional Classification, Korean version of modified Barthel index, and grip strength.Results: At 3 weeks and 3 months, the comparison of change scores in KOVAL between two groups revealed difference of 0.84 (95% CI: -1.19, -0.49; p for trend = .000) and 1.21 (95% CI: -2.05, -0.36; p for trend = .006), respectively. At 3 weeks, comparison of change score in FAC between two groups revealed a difference of 0.73 (95% CI: 0.28, 1.19; p for trend = .003). The comparison of change scores between two groups also showed a difference in the 6 months in KOVAL and in the 3 and 6 months in FAC. The comparison of changes in scores in BBS between two groups revealed difference of 11.63 (95% CI: 5.85, 17.40; p for trend = .001), 9.00 (95% CI: 2.28, 15.71; p for trend = .006), and 11.05 (95% CI: 3.62, 18.48; p for trend = .006), respectively, at each follow-up.Conclusions: Both groups were improved after intervention. As additional benefits were evident among those who carried out AGT, it may be appropriate for patients with sarcopenia after hip fracture surgery. [ABSTRACT FROM AUTHOR]

Key–value data structures have been extensively used in various applications. When a large amount of data needs to be compactly stored in a fixed memory size, a functional Bloom filter is a space-efficient key–value structure. In this letter, we propose a 2-stage functional Bloom filter structure composed of a primary functional Bloom filter and a secondary functional Bloom filter to resolve the indeterminables produced from the primary functional Bloom filter. We analytically present the memory ratio allocated for each of the two Bloom filters to achieve the lowest search failure rate. The analytical result is validated through experiments, thereby demonstrating that the optimal performance is realized when the secondary functional Bloom filter uses 3% of the total memory. [ABSTRACT FROM AUTHOR]

Hash-based data structures have been widely used in many applications. An intrinsic problem of hashing is collision, in which two or more elements are hashed to the same value. If a hash table is heavily loaded, more collisions would occur. Elements that could not be stored in a hash table because of the collision cause search failures. Many variant structures have been studied to reduce the number of collisions, but none of the structures completely solves the collision problem. In this paper, we claim that a functional Bloom filter (FBF) provides a lower search failure rate than hash tables, when a hash table is heavily loaded. In other words, a hash table can be replaced with an FBF because the FBF is more effective than hash tables in the search failure rate in storing a large amount of data to a limited size of memory. While hash tables require to store each input key in addition to its return value, a functional Bloom filter stores return values without input keys, because different index combinations according to each input key can be used to identify the input key. In search failure rates, we theoretically compare the FBF with hash-based data structures, such as multi-hash table, cuckoo hash table, and d-left hash table. We also provide simulation results to prove the validity of our theoretical results. The simulation results show that the search failure rates of hash tables are larger than that of the functional Bloom filter when the load factor is larger than 0.6. [ABSTRACT FROM AUTHOR]

Objective: The aim of this study was to investigate the clinical implications of diabetes for the management of idiopathic sudden sensorineural hearing loss (ISSNHL).Study Design: Retrospective study.Setting: Tertiary referral center.Subjects and Methods: ISSNHL patients (N = 403) who received inpatient management between January 2015 and December 2018 were analyzed. All were managed by a uniform treatment protocol of high-dose steroid therapy and salvage intratympanic steroid injections. Treatment results were evaluated according to the American Academy of Otolaryngology-Head and Neck Surgery's criteria 3 months after the start of treatment. We compared the clinical parameters and treatment outcomes of ISSNHL with and without diabetes. We also evaluated the influence of diabetes on the prognosis of ISSNHL by propensity score matching.Results: Overall, of the 403 ISSNHL patients, 94 (23.3%) had diabetes, and 11 were newly diagnosed with diabetes. The patients with diabetes were older than those without diabetes (P < .001), and their initial hearing threshold was significantly higher (P < .001). The diabetic patients were hospitalized for a longer period, and their hearing recovery rate was lower. However, when age, sex, and initial hearing level were adjusted by propensity score matching, the diabetic patients and matched controls yielded similar treatment results.Conclusions: ISSNHL with diabetes usually presents with severe hearing loss and requires longer hospitalization. However, diabetes itself may not influence the prognosis of ISSNHL. Proper management must be provided in ISSNHL with diabetes. [ABSTRACT FROM AUTHOR]

Background: Vestibular schwannoma (VS) is a benign intracranial neoplasm originating in the Schwann cells of the vestibular nerve. Despite its origin, the most common symptom is sensorineural hearing loss which is presented in more than 90% of patients. The underlying pathophysiology of this hearing loss has not been fully understood.Objective: To assess the in vivo function of cochlear inner hair cells and spiral ganglion neurons in VS, cochlear dead regions (DRs) were evaluated via the threshold-equalizing noise (TEN) test in untreated VS patients.Method: Untreated patients diagnosed with sporadic unilateral VS and normal contralesional hearing were enrolled from July 2011 to June 2016. Audiometric evaluation including TEN tests were performed. Based on the magnetic resonance findings, characteristics of individual tumors were assessed.Results: The average pure-tone threshold (word recognition score [WRS]) of 23 enrolled patients was 42.7 dB (76.1%). Nineteen DRs (11.8% of 161 tested frequencies) were found in 8 patients (34.8% of enrolled cases). Among the intracanalicular (IAC) tumors, 6 out of 10 ears (60%) carried DRs, while 2 of 13 (15.4%) showed DRs among the cerebellopontine angle (CPA) lesions (p = 0.039). Pure-tone thresholds and WRS were not different between the two groups. Logistic regression analysis showed that the tumor location, IAC versus CPA, was significantly associated with DRs (p = 0.041, Nagelkerke R2 = 0.471), whereas age, sex, tumor size, distance from the tumor to the cochlea, T2-weighted hypointensity on the MRI and pure-tone thresholds showed no significance.Conclusions: Cochlear DRs are detected in hearing losses associated with unilateral sporadic VS using the TEN test. Individual DRs were detected variously in high, mid, or low frequencies. In our preliminary data, IAC tumors showed a higher number of DRs than CPA tumors despite similar average hearing thresholds. Further studies including longitudinal follow-up of hearing as well as change in DRs may provide useful information about VS patients. [ABSTRACT FROM AUTHOR]

The Internet Protocol (IP) address lookup is one of the most challenging tasks for Internet routers, since it requires to perform packet forwarding at wire-speed for tens of millions of incomming packets per second. Efficient IP address lookup algorithms have been widely studied to satisfy this requirement. Among them, Bloom filter-based approach is attractive in providing high performance. This paper proposes a high-speed and flexible architecture based on a vectored-Bloom filter (VBF), which is a space-efficient data structure that can be stored in a fast on-chip memory. An off-chip hash table is infrequently accessed, only when the VBF fails to provide address lookup results. The proposed architecture has been evaluated through both a behavior simulation with C language and a timing simulation with Verilog. The hardware implementation result shows that the proposed architecture can achieve the throughput of 5 million packets per second in a field programmable gate array (FPGA) operated at 100 MHz. [ABSTRACT FROM AUTHOR]

Objective: The aim of this study was to investigate the clinical significance of 4‐hour delayed‐enhanced 3.0 Tesla (3T) 3D‐fluid‐attenuated inversion recovery (FLAIR) MR imaging in sudden sensorineural hearing loss (SSNHL). Study design: Case series with comparisons. Setting: Tertiary referral centre. Participants: Eighty‐seven idiopathic SSNHL patients were enrolled between January 2015 and December 2016 and received high dose steroid therapy and intratympanic steroid injections as salvage treatment. Intervention: Pre‐contrast, 10‐minute and 4‐hour delayed‐enhanced 3D‐FLAIR MR images were obtained using double‐dose IV gadolinium. Main outcome measures: The results of treatment were evaluated according to Siegel's criteria 3 months after the start of treatment. Where possible lesion‐side laterality of the inner ear was identified based on the MR images, the associations between MR findings and other clinical parameters were analysed, and the relationships between hearing recovery and MR image findings were assessed. Results: Lesion‐side laterality was identified on MRI in 52 (59.7%), 18 (20.1%) and 8 (9.2%) patients, based on 4‐hour delayed, 10‐minute delayed, and pre‐contrast images, respectively. The hearing recovery rate was significantly lower in the patients with lesion‐side laterality on 4‐hour delayed images (P < 0.001). In a multivariate analysis, lesion‐side laterality on 4‐hour delayed images was associated with poor prognosis (OR = 5.6) after adjusting other prognostic factors including initial hearing level, lesion‐side laterality on 10‐min delayed images and presence of vertigo. In addition, as the extent of enhancement in the inner ear increased the probability of hearing recover decreased (P = 0.001). Conclusions: Contrast enhancement of inner ear structures can be seen on 4‐hour delayed‐enhanced 3T 3D‐FLAIR MR images in idiopathic SSNHL. Asymmetric lesion‐side enhancement of the inner ear may be associated with a poor prognosis. [ABSTRACT FROM AUTHOR]

Network traffic has increased rapidly in recent years, mainly associated with the massive growth of various applications on mobile devices. Named data networking (NDN) technology has been proposed as a future Internet architecture for effectively handling this ever-increasing network traffic. In order to realize the NDN, high-speed lookup algorithms for a forwarding information base (FIB) are crucial. This paper proposes a level-priority trie (LPT) and a 2-phase Bloom filter architecture implementing the LPT. The proposed Bloom filters are sufficiently small to be implemented with on-chip memories (less than 3 MB) for FIB tables with up to 100,000 name prefixes. Hence, the proposed structure enables high-speed FIB lookup. The performance evaluation result shows that FIB lookups for more than 99.99% of inputs are achieved without needing to access the database stored in an off-chip memory. [ABSTRACT FROM AUTHOR]

Objective: To investigate the clinical significance of 4-hour delayed-enhanced 3.0 Tesla three-dimensional (3D) fluid-attenuated inversion recovery (FLAIR) magnetic resonance (MR) imaging in acute vestibular neuritis.Study Design: A prospective observational study.Methods: Twenty-nine vestibular neuritis patients were enrolled between January 2017 and June 2017. Vestibular function tests, comprising the caloric and video head impulse tests and vestibular-evoked myogenic potential measurements, were performed. Precontrast, 10-minute, and 4-hour delayed-enhanced 3D-FLAIR MR images using double-dose IV gadolinium were obtained. After laterality and extent of inner ear enhancement were defined, the patients were divided into groups based on the patterns of enhancement, and clinical parameters were analyzed according to the groups.Results: Twenty patients (20 of 29, 69.0%) had obviously asymmetric enhancement of the affected inner ear structures on 4-hour delayed images, whereas only three patients (10.3%) had marked enhancement on 10-minute delayed images. The duration of spontaneous nystagmus (DurSN) was significantly longer in the patients with enhancement, especially with enhancement of the whole inner ear, including the vestibule and semicircular canals (P < 0.033). Spontaneous nystagmus resolved within 12 days in patients without laterality of enhancement, and within 16 days in ipsilesional enhancement confined to the inner auditory canal and fundus. Other results of vestibular function tests did not reveal any significant associations with MR enhancement.Conclusions: Contrast enhancement of the vestibular nerve and inner ear structures can be identified on 4-hour delayed-enhanced 3T 3D-FLAIR MR images in acute vestibular neuritis. The extent of inner ear enhancement may be associated with the DurSN.Level Of Evidence: 4. Laryngoscope, 1946-1951, 2018. [ABSTRACT FROM AUTHOR]

A counting Bloom filter (CBF) is commonly used in many applications for the membership queries of dynamic data since the CBF can provide delete operations. A CBF uses an array of $c$ -bit counters. The $c$ should be large enough to avoid overflows. In this letter, we propose an alternative to CBF, named ternary Bloom filter (TBF) for performance improvement. The proposed TBF allocates the minimum number of bits to each counter and includes more number of counters instead to reduce false positive probability. We present a mathematical analysis and experimental results for a set of performance measures. When the TBF consumes the same amount of memory as the CBF, the TBF provides much lower false positive rates than the CBF. [ABSTRACT FROM PUBLISHER]

Objectives/Hypothesis To evaluate the prognostic and predictive value of electroneuronography (ENoG) in acute severe inflammatory facial paralysis, including Bell's palsy and Ramsay Hunt syndrome ( RHS). Study Design Prospective observational study. Methods Patients with acute severe facial paralysis of House-Brackmann (H-B) grade IV or worse and diagnosed with Bell's palsy or RHS were enrolled from August 2007 to July 2011. After treatment with oral corticosteroid, antiviral agent, and protective eye care, patients were followed up until recovery or 12 months from onset. Results Sixty-six patients with Bell's palsy and 22 with RHS were included. Multiple logistic regression analysis showed significant effect of ENoG value on recovery in both Bell's palsy and RHS. Values of ENoG were significantly worse in RHS than Bell's palsy. Chance of early recovery within 6 weeks after correction of ENoG effect was still significantly worse in RHS. Logistic regression analysis showed 90% chance of recovery within 6 months, expected with ENoG values of 69.2% degeneration (Bell's palsy) and 59.3% ( RHS). The receiver operating characteristics ( ROC) curves showed ENoG values of 82.5% (Bell's palsy) and 78.0% ( RHS) as a critical cutoff value of nonrecovery until 1 year, with the best sensitivity and specificity. Conclusions A higher chance of recovery was expected with better ENoG in Bell's palsy and RHS. Based on our data, nonrecovery is predicted in patients with ENoG value greater than 82.5% in Bell's palsy, and 78% in RHS. Level of Evidence 4. Laryngoscope, 123:2526-2532, 2013 [ABSTRACT FROM AUTHOR]

Cochlear synaptopathy refers to a subclinical hearing pathology which could potentially explain hearing difficulties within the normal hearing threshold; it is also called "hidden hearing loss". We hypothesized that a temporary threshold shift in sudden sensorineural hearing loss (ISSNHL) also affects the function in the synapse. The aim of the study was to evaluate the presence of cochlear synaptopathy in patients who had completely recovered from unilateral SSNHL Nineteen patients who had completely recovered from ISSNHL from January 2018 to June 2021 were assessed. Complete recovery was established by pure tone audiometry (PTA) 3 months after treatment, according to the American Academy of Otolaryngology–Head and Neck Surgery criteria. Subjects completed the pure tone audiometry, speech audiometry and auditory brain stem response (ABR) test, and completed a questionnaire regarding hearing loss after hearing recovery. The ABR amplitudes of wave I and wave V, and the ratio of wave I/V of both ears (recovered side and healthy side) were assessed. A visual analog scale (VAS) and a hidden hearing loss questionnaire were used to evaluate subjective hearing difficulty. The ABR waves I of the recovered ears had a significantly lower amplitude (p = 0.002) than those of the healthy side, whereas there was no difference in wave V (p = 0.985) or in the ratio of wave I/V (p = 0.107). Some patients still felt mild hearing difficulty although their PTA results were normal, but there was no clear relationship between the VAS score, wave I amplitude and speech recognition scores. The present findings point to the possible existence of cochlear synaptopathy in ears that have completely recovered from unilateral sudden sensorineural hearing loss. We suggest that the causes of cochlear synaptopathy and of idiopathic sudden hearing loss may have something in common. [ABSTRACT FROM AUTHOR]

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. [ABSTRACT FROM AUTHOR]