A deep learning approach for detecting otitis media from wideband tympanometry measurements

Factors affecting the antimicrobial changes during treatment for acute otitis media in Japan: A retrospective cohort study using classification and regression trees (CART) analysis

OBJECTIVES

Factors that affect the change of first-line antimicrobial agents were investigated to further promote their appropriate use.

METHODS

This descriptive study used an electronic medical records database. Total 16,353 of the 199,896 patients enrolled between 1996 and 2019 met the inclusion criteria and formed the overall pediatric acute otitis media (AOM) cohort. The factors leading to the change in first-line antimicrobial agents within 14 days were analyzed using classification and regression trees (CART) analysis.

RESULTS

This antimicrobial treatment cohort, involved 4860 cases of AOM alone and 9567 cases of AOM with other diseases. The size of the medical facility based on number of beds and historical duration of patient registration impacted on antimicrobial changes.

CONCLUSIONS

The current results show that hospital-wide or nation-wide antimicrobial stewardship promotion could be the most influencing factor for antimicrobial changes. Particularly in cases of AOM where other diseases coexist, a more accurate diagnosis and definition of treatment failure of first-line drug are suggested to be important while establishing future treatment strategies. The current study is important to promote appropriate antimicrobial use for AOM treatment.

Preliminary results of classifying otosclerosis and disarticulation using a convolutional neural network trained with simulated wideband acoustic immittance data

Current noninvasive methods of clinical practice often do not identify the causes of conductive hearing loss due to pathologic changes in the middle ear with sufficient certainty. Wideband acoustic immittance (WAI) measurement is noninvasive, inexpensive and objective. It is very sensitive to pathologic changes in the middle ear and therefore promising for diagnosis. However, evaluation of the data is difficult because of large interindividual variations. Machine learning methods like Convolutional neural networks (CNN) which might be able to deal with this overlaying pattern require a large amount of labeled measurement data for training and validation. This is difficult to provide given the low prevalence of many middle-ear pathologies. Therefore, this study proposes an approach in which the WAI training data of the CNN are simulated with a finite-element ear model and the Monte-Carlo method. With this approach, virtual populations of normal, otosclerotic, and disarticulated ears were generated, consistent with the averaged data of measured populations and well representing the qualitative characteristics of individuals. The CNN trained with the virtual data achieved for otosclerosis an AUC of 91.1 %, a sensitivity of 85.7 %, and a specificity of 85.2 %. For disarticulation, an AUC of 99.5 %, sensitivity of 100 %, and specificity of 93.1 % was achieved. Furthermore, it was estimated that specificity could potentially be increased to about 99 % in both pathological cases if stapes reflex threshold measurements were used to confirm the diagnosis. Thus, the procedures' performance is comparable to classifiers from other studies trained with real measurement data, and therefore the procedure offers great potential for the diagnosis of rare pathologies or early-stages pathologies. The clinical potential of these preliminary results remains to be evaluated on more measurement data and additional pathologies.

Toward Automating Diagnosis of Middle- and Inner-ear Mechanical Pathologies With a Wideband Absorbance Regression Model

OBJECTIVES

During an initial diagnostic assessment of an ear with normal otoscopic exam, it can be difficult to determine the specific pathology if there is a mechanical lesion. The audiogram can inform of a conductive hearing loss but not the underlying cause. For example, audiograms can be similar between the inner-ear condition superior canal dehiscence (SCD) and the middle-ear lesion stapes fixation (SF), despite differences in pathologies and sites of lesion. To gain mechanical information, wideband tympanometry (WBT) can be easily performed noninvasively. Absorbance , the most common WBT metric, is related to the absorbed sound energy and can provide information about specific mechanical pathologies. However, absorbance measurements are challenging to analyze and interpret. This study develops a prototype classification method to automate diagnostic estimates. Three predictive models are considered: one to identify ears with SCD versus SF, another to identify SCD versus normal, and finally, a three-way classification model to differentiate among SCD, SF, and normal ears.

DESIGN

Absorbance was measured in ears with SCD and SF as well as normal ears at both tympanometric peak pressure (TPP) and 0 daPa. Characteristic impedance was estimated by two methods: the conventional method (based on a constant ear-canal area) and the surge method, which estimates ear-canal area acoustically.Classification models using multivariate logistic regression predicted the probability of each condition. To quantify expected performance, the condition with the highest probability was selected as the likely diagnosis. Model features included: absorbance-only, air-bone gap (ABG)-only, and absorbance+ABG. Absorbance was transformed into principal components of absorbance to reduce the dimensionality of the data and avoid collinearity. To minimize overfitting, regularization, controlled by a parameter lambda, was introduced into the regression. Average ABG across multiple frequencies was a single feature.Model performance was optimized by adjusting the number of principal components, the magnitude of lambda, and the frequencies included in the ABG average. Finally, model performances using absorbance at TPP versus 0 daPa, and using the surge method versus constant ear-canal area were compared. To estimate model performance on a population unknown by the model, the regression model was repeatedly trained on 70% of the data and validated on the remaining 30%. Cross-validation with randomized training/validation splits was repeated 1000 times.

RESULTS

The model differentiating between SCD and SF based on absorbance-only feature resulted in sensitivities of 77% for SCD and 82% for SF. Combining absorbance+ABG improved sensitivities to 96% and 97%. Differentiating between SCD and normal using absorbance-only provided SCD sensitivity of 40%, which improved to 89% by absorbance+ABG. A three-way model using absorbance-only correctly classified 31% of SCD, 20% of SF and 81% of normal ears. Absorbance+ABG improved sensitivities to 82% for SCD, 97% for SF and 98% for normal. In general, classification performance was better using absorbance at TPP than at 0 daPa.

CONCLUSION

The combination of wideband absorbance and ABG as features for a multivariate logistic regression model can provide good diagnostic estimates for mechanical ear pathologies at initial assessment. Such diagnostic automation can enable faster workup and increase efficiency of resources.

Multi-modal deep learning for joint prediction of otitis media and diagnostic difficulty

Objectives

In this study, we propose a diagnostic model for automatic detection of otitis media based on combined input of otoscopy images and wideband tympanometry measurements.

Methods

We present a neural network-based model for the joint prediction of otitis media and diagnostic difficulty. We use the subclassifications acute otitis media and otitis media with effusion. The proposed approach is based on deep metric learning, and we compare this with the performance of a standard multi-task network.

Results

The proposed deep metric approach shows good performance on both tasks, and we show that the multi-modal input increases the performance for both classification and difficulty estimation compared to the models trained on the modalities separately. An accuracy of 86.5% is achieved for the classification task, and a Kendall rank correlation coefficient of 0.45 is achieved for difficulty estimation, corresponding to a correct ranking of 72.6% of the cases.

Conclusion

This study demonstrates the strengths of a multi-modal diagnostic tool using both otoscopy images and wideband tympanometry measurements for the diagnosis of otitis media. Furthermore, we show that deep metric learning improves the performance of the models.

Study on characteristics of wideband acoustic immittance in patients with Inner Ear Malformations

OBJECTIVE

To compare the differences in wideband absorbance and the resonance frequency (RF) between patients with inner ear malformations and normal control, and to explore the auditory diagnostic value of wideband acoustic immittance (WAI).

METHODS

A total of 38 patients (59 ears) with enlarged vestibular aqueduct (EVA), 13 patients (14 ears) with incomplete partition type I (IP-I) and 13 patients (26 ears) with incomplete partition type II (IP-II) were included. 50 normal control (100 ears). All subjects underwent WAI tests to compare the absorbance configuration and resonance frequency.

RESULTS

All the group showed lower absorbance at ambient pressure than at peak pressure in certain frequencies under 2000Hz. Under 1000Hz, the absorbance of EVA was higher than that of other groups. The average absorbance and highest absorbance of IP-I were the lowest(P＜0.05). However, IP-II and normal group had similarity on some characteristics. The three IEM groups mainly different at low and high frequencies, but not at medium frequencies. The highest absorbance of all the groups were appeared around 3000Hz. The RF of all the groups from low to high were EVA＜IP-II＜normal control＜IP-I, and the lowest was EVA(P＜0.05).

CONCLUSION

Inner ear malformations can affect energy absorbance and RF. WAI is sensitive and non-invasive to provide useful information about inner ear status and facilitate detection of ear pathology.

Diagnosis, Treatment, and Management of Otitis Media with Artificial Intelligence

A common infectious disease, otitis media (OM) has a low rate of early diagnosis, which significantly increases the difficulty of treating the disease and the likelihood of serious complications developing including hearing loss, speech impairment, and even intracranial infection. Several areas of healthcare have shown great promise in the application of artificial intelligence (AI) systems, such as the accurate detection of diseases, the automated interpretation of images, and the prediction of patient outcomes. Several articles have reported some machine learning (ML) algorithms such as ResNet, InceptionV3 and Unet, were applied to the diagnosis of OM successfully. The use of these techniques in the OM is still in its infancy, but their potential is enormous. We present in this review important concepts related to ML and AI, describe how these technologies are currently being applied to diagnosing, treating, and managing OM, and discuss the challenges associated with developing AI-assisted OM technologies in the future.

Three-dimensional wideband absorbance immittance findings in young adults with large vestibular aqueduct syndrome

Objective

To investigate the effect of large vestibular aqueduct syndrome (LVAS) on middle ear sound transmission using wideband absorbance immittance (WAI).

Methods

WAI results from young adult LVAS patients and normal adults were compared.

Results

Averaged energy absorbance (EA) at ambient and peak pressure in the LVAS group showed differences to the normal group. Under ambient pressure, the average EA of the LVAS group was significantly higher than the normal group at frequencies 472-866 Hz and 6169-8000 Hz (p < .05) and lower at frequencies 1122-2520 Hz (p < .05). Under peak pressure, absorbance was increased at frequencies 515-728, 841, and 6169-8000 Hz (p < .05) and decreased at 1122-1374 Hz and 1587-2448 Hz (p < .05). An investigation into the effect of external auditory canal pressure on EA across frequencies in the pressure-frequency domain, showed that EA differed significantly in the low-frequency region of 707and 1000 Hz from 0 to 200 daPa and 500 Hz at 50 daPa (p < .05). There was also a significant difference in EA between the two groups at 8000 Hz (p < .05) in the pressure range -200-300 daPa.

Conclusion

WAI is a valuable tool to measure the effect of LVAS on middle ear sound transmission. LVAS has a significant effect on EA at low and mid frequencies under ambient pressure, while the frequencies affected are mainly at low frequencies when positive pressure is presented.

Level of Evidence

Level 3a.