SFI Method Shows High Accuracy in Predicting Middle-Ear Dysfunctions

Clinical Question

How effective is a machine-based learning method at predicting middle-ear dysfunctions?

Bottom Line

Study results suggest that the sweep frequency impedance (SFI) method will lead to more accurate predictions of middle-ear dysfunctions in clinical settings as compared to conventional tympanometry.

Background: The SFI meter is an apparatus that delivers a frequency-sweeping sound into the ear canal and evaluates dynamic characteristics of the middle ear based on sound pressure changes. The measurement of acoustic impedance in the ear canal is considered a valid method for indirectly monitoring middle-ear conditions.

Study design: Investigative study

Setting: Faculty of Frontier Engineering, Graduate School Department, University of Yamanashi, Kofu, Japan

Synopsis: For their study, researchers used a renewed SFI meter, consisting of a personal computer, an audio interface, and a probe system, to detect sound pressure variations with a higher signal-to-noise ratio. They conducted the SFI measurement with normal and impaired ears using the renewed meter and proposed a machine learning-based prediction method of middle-ear dysfunctions in two-dimensional characteristics. Measurements were conducted on 74 ears of 53 participants (28 males), including 42 ears of 25 participants with normal hearing function, 18 ears of 14 patients with middle-ear fixation, and 10 ears of 10 patients with middle-ear separation. Results showed that ossicular chain fixation was predicted with an accuracy of 0.80 and an area under the receiver operating characteristic curve (AUC) of 0.86, and ossicular chain separation with an accuracy of 1.0 and an AUC of 1.0. When only impaired ears were analyzed as the classification targets, the team’s proposed method completely classified the fixation and separation ears. The authors say findings suggest that the SFI method will lead to more accurate predictions of middle-ear dysfunctions in clinical settings than conventional tympanometry. Study limits included limited data on proximity values, PF  (predict fixation), and PS (predict separation).

Citation: Toya T, et al. Machine-learning-based classification of middle-ear fixation and separation using sweep frequency impedance information reflecting middle-ear dynamics. J Acoust Soc Am. 2025;157:3624-3637. doi:10.1121/10.0036639.