Inter-rater reliability of the diagnosis of otitis media based on otoscopic images and wideband tympanometry measurements

Sundgaard, Josefine Vilsbøll; Värendh, Maria; Nordström, Franziska U; Kamide, Yosuke; Tanaka, Chiemi; Harte, James M.; Paulsen, Rasmus Reinhold; Christensen, Anders Nymark; Bray, Peter; Laugesen, Søren

doi:10.1016/j.ijporl.2021.111034

Cited by 6 publications

(5 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, each ENT rated their certainty of each diagnosis on the scale: very low, low, medium, moderate, or high, which is converted to a scale from 0 to 1. More details on this dataset are in Sundgaard et al [28]. For a case, μ correct is the fraction of correct ENT answers and μ certainty is the average self-evaluated certainty.…”

Section: Datasetsmentioning

confidence: 99%

Was that so Hard? Estimating Human Classification Difficulty

Hannemose

Sundgaard

Ternov

et al. 2022

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

When doctors are trained to diagnose a specific disease, they learn faster when presented with cases in order of increasing difficulty. This creates the need for automatically estimating how difficult it is for doctors to classify a given case. In this paper, we introduce methods for estimating how hard it is for a doctor to diagnose a case represented by a medical image, both when ground truth difficulties are available for training, and when they are not. Our methods are based on embeddings obtained with deep metric learning. Additionally, we introduce a practical method for obtaining ground truth human difficulty for each image case in a dataset using self-assessed certainty. We apply our methods to two different medical datasets, achieving high Kendall rank correlation coefficients on both, showing that we outperform existing methods by a large margin on our problem and data.

show abstract

Section: Datasetsmentioning

confidence: 99%

Was that so Hard? Estimating Human Classification Difficulty

Hannemose

Sundgaard

Ternov

et al. 2022

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

show abstract

“…The multi‐modal model performs better for both classification and difficulty estimation compared to the models trained on the two modalities separately. The four ENTs in the human inter‐rater study by Sundgaard et al 24 achieved 64% accuracy on this dataset based on the same amount of patient information used in the multi‐modal embedding model, which achieved 86%. This substantial increase in performance is very promising for a future diagnostic tool and shows the strength of deep learning models for medical image analysis.…”

Section: Discussionmentioning

confidence: 96%

“…More details on the human inter‐rater study with the four ENTs can be found in Sundgaard et al 24 The ground truth diagnosis used for training the models for this paper is based solely on the diagnosis by the original ENT, who assessed the patients in the clinic. There is, of course, a risk of human errors in the annotation process, but no other solution for achieving a ground truth diagnosis was possible in the current study.…”

Section: Methodsmentioning

confidence: 99%

“…More details on the human inter-rater study with the four ENTs can be found in Sundgaard et al 24 The ground truth diagnosis used As seen in Table 1, the highest classification performance is achieved by the combined image and WBT model, as both accuracy and each class-wise F1-score are superior to the scores obtained with the other methods. The results also show that Kendall's τ for difficulty estimation is increased when the model is trained on both images and WBT measurements.…”

Section: Datamentioning

confidence: 99%

See 1 more Smart Citation

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

Sundgaard,

Hannemose,

Laugesen

et al. 2024

Laryngoscope Investig Oto

Self Cite

View full text Add to dashboard Cite

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

show abstract

“…Due to the VA enlargement, the volume of the inner ear increases. This reduces the conduction of high-frequency sound ( Sundgaard et al, 2022 ), i.e., a decline in absorbance, suggesting that an increased inner ear volume should be part of the mass in the whole sound transmission system. From the audiology perspective, LVAS often manifests as characteristic low-frequency air-bone conduction differences, but no reports are available on air-bone conduction differences in patients with high-frequency residual hearing, probably suggesting a difference in the absorbance of low-frequency and high-frequency sounds.…”

Section: Discussionmentioning

confidence: 99%

Characteristics of large vestibular aqueduct syndrome in wideband acoustic immittance

Gao³

et al. 2023

Front. Neurosci.

View full text Add to dashboard Cite

ObjectiveTo describe the characteristics of large vestibular aqueduct syndrome (LVAS) in wideband acoustic immittance (WAI) and to explore whether inner ear deformity has an impact on WAI results.MethodsSubjects with typical LVAS (LVAS group) and control subjects with a normal anatomical structure of the inner ear (control group) were screened from pediatric patients with cochlear implants using thin-slice computed tomography (CT) images of the temporal bone. With inflammation of the auditory canal and middle ear excluded by routine ear examination and 226 Hz acoustic immittance, WAI data were acquired. Then, the maximum absorbance as the major observation indicator on the mean tympanogram was compared between the LVAS group and control group, and a descriptive comparison of the mean tympanogram and frequency-absorbance curve at peak pressure was performed between the two groups.ResultsThe LVAS group included 21 cases (38 ears), and the control group included 27 cases (45 ears). All LVAS subjects met the Valvassori criteria, and the VA at the horizontal semicircular canal displayed flared expansion. On the mean tympanogram, the maximum absorbance in the LVAS group (0.542 ± 0.087) was significantly higher than that in the control group (0.455 ± 0.087) (p < 0.001). The tympanogram in the LVAS group showed an overall elevation, and the absorbance at all pressure sampling points was significantly higher than that in the control group (p < 0.001). The frequency-absorbance curve at peak pressure first increased and then decreased in both groups, and the LVAS group showed higher absorbance than the control group in the frequency range below 2,828 Hz. The absorbance at 343–1,124 Hz was significantly different between the two groups (p < 0.001), and 343–1,124 Hz was the major frequency range at which the maximum absorbance on the mean tympanogram increased in the LVAS group.ConclusionLarge vestibular aqueduct syndrome (LVAS) shows increased absorbance in low and medium frequency ranges in WAI. The maximum absorbance on the mean tympanogram can serve as a reliable evaluation indicator. Inner ear factors must be considered when middle ear lesions are analyzed by WAI.

show abstract

Inter-rater reliability of the diagnosis of otitis media based on otoscopic images and wideband tympanometry measurements

Cited by 6 publications

References 17 publications

Was that so Hard? Estimating Human Classification Difficulty

Was that so Hard? Estimating Human Classification Difficulty

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

Characteristics of large vestibular aqueduct syndrome in wideband acoustic immittance

Contact Info

Product

Resources

About