An objective method for detecting tinnitus pitch based on the fine structure of otoacoustic emissions at stimulation frequency

Abstract

This invention discloses an objective detection method for tinnitus pitch based on the fine structure of otoacoustic emissions (SFOAE) at stimulation frequency, belonging to the field of audiological testing technology. The method includes: acquiring the fine structure signal of SFOAE and PTA hearing threshold data from the tinnitus ear of the subject; performing multi-level threshold feature extraction on the SFOAE signal to obtain primary feature valley points, secondary feature valley points, and weighted adjustment points; calculating the SFOAE amplitude normalization weight and hearing threshold mapping weight based on the feature point set, and performing a two-factor composite weighted calculation to obtain a preliminary predicted frequency; identifying and secondarily correcting the significant fluctuation areas of the preliminary prediction result to obtain a secondary corrected frequency; and optimizing the prediction result based on a nearest feature point conditional fusion strategy to obtain the final predicted frequency. This invention can objectively and accurately predict tinnitus frequency without requiring subjective feedback from the patient, providing a reliable data foundation for the development of personalized sound therapy plans.

Description

Technical Field

[0001] This invention relates to the field of audiological testing technology, specifically to an objective method for detecting tinnitus pitch based on the fine structure of otoacoustic emissions at stimulation frequency. Background Technology

[0002] Tinnitus is a subjective hearing disorder in which patients perceive sounds in their ears or skull even without an external sound source. Objective diagnosis and accurate frequency matching are major challenges in clinical audiology research. Currently, determining tinnitus frequency in clinical practice mainly relies on the subjective behavioral fitting method, which involves having patients compare external sound stimuli with their subjective tinnitus perception, gradually adjusting the stimulus frequency until the patient confirms a match.

[0003] However, existing subjective matching methods have inherent limitations: approximately 25% of patients experience octave confusion during the initial matching, mistaking adjacent octaves of the actual tinnitus frequency for matching tones. This causes the notch center frequency to deviate completely from the lesion area during sound therapy, resulting in a loss of treatment effectiveness. Furthermore, the test-retest reliability of subjective matching results is poor, with significant fluctuations in results from repeated measurements, reflecting the dynamic nature of tinnitus perception itself and the inherent cognitive noise in the subjective matching process. In addition, subjective matching methods are completely unsuitable for groups with cognitive impairments or other conditions that prevent them from providing accurate subjective feedback.

[0004] On the other hand, existing methods for objectively detecting tinnitus frequencies are relatively scarce. Previous studies have found a correlation between tinnitus frequencies and the minimum points of the fine structure of the most nearest neighbor echolocation (SFOAE) in the nearest frequency domain. However, current technologies lack integrated objective detection methods, making it difficult to accurately assess tinnitus frequencies point by point, thus hindering objective and accurate prediction of tinnitus frequencies. Furthermore, existing otoacoustic emission (OAE) detection devices typically use single-shot acquisition or simple superposition methods when acquiring SFOAE signals, resulting in insufficient signal-to-noise ratios. Moreover, the sweep parameters are not optimized for tinnitus detection scenarios, leading to insufficient quality and resolution of the SFOAE fine structure signals to meet the requirements for accurate tinnitus frequency prediction. Simultaneously, existing technologies lack a systematic extraction and multi-level screening mechanism for SFOAE fine structure feature points, making it difficult to effectively distinguish feature points with different confidence levels and integrate multi-dimensional information such as hearing thresholds for comprehensive judgment.

[0005] Therefore, there is an urgent need for a detection method that can objectively and accurately predict tinnitus frequency to overcome the shortcomings of existing technologies. Summary of the Invention

[0006] To address the problems of existing technologies where tinnitus frequency matching is highly dependent on patient subjective feedback, results are highly volatile, and are not applicable to specific populations, this invention provides an objective detection method for tinnitus pitch based on the fine structure of otoacoustic emissions at stimulation frequencies, overcoming the shortcomings of existing technologies.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] This invention provides an objective method for detecting tinnitus pitch based on the fine structure of otoacoustic emissions at stimulation frequency, comprising:

[0009] Step S1: Collect the otoacoustic emission fine structure signal of the stimulation frequency of the subject's tinnitus ear and the pure tone audiometry threshold data;

[0010] Step S2: Perform multi-level threshold feature extraction on the otoacoustic emission fine structure signal at the stimulation frequency to obtain the set of first-level feature valley points, the set of second-level feature valley points, and the set of weighted adjustment points;

[0011] Step S3: Based on the set of first-level feature valley points and the set of second-level feature valley points, calculate the normalized weight of the stimulation frequency otoacoustic emission amplitude and the mapping weight of the hearing threshold respectively, and perform a two-factor composite weighted calculation of stimulation frequency otoacoustic emission-pure tone audiometry to obtain the preliminary predicted frequency;

[0012] Step S4: Based on the set of weight adjustment points, identify and correct the significant fluctuation areas of the obtained preliminary prediction results.

[0013] Step S5: Based on the nearest feature points selected from the extracted first-level feature valley point set and second-level feature valley point set, perform conditional fusion optimization on the prediction results to obtain the final prediction frequency;

[0014] Step S6: Based on the entire prediction process described above, and depending on whether the data has undergone secondary correction and condition fusion, set the prediction frequency obtained in the last step as the final prediction frequency. .

[0015] Preferably, the acquisition process of the stimulation frequency otoacoustic emission fine structure signal in step S1 is as follows: a stimulation signal using an exponential sweep frequency mode is employed, with a lower sweep frequency of... The upper limit frequency is During testing, the frequency ratio of the suppressor tone to the probe tone was strictly controlled to a fixed value, the probe tone intensity was set within a reasonable testing range, and testing was conducted according to the set testing step size. The suppressor tone intensity and the probe tone maintained a fixed dynamic relationship, and the duration of a single frequency sweep was set to [missing value]. Furthermore, the data is averaged N times during the acquisition process to suppress environmental and physiological background noise and improve the signal-to-noise ratio.

[0016] The process of collecting pure-tone audiometry threshold data is as follows: using air conduction pure-tone audiometry, the hearing threshold is determined using the rise and fall method, and the air conduction hearing threshold of the subject is recorded at N standard frequency points from 125Hz to 16000Hz, where N≥2.

[0017] Preferably, step S2 specifically includes the following steps:

[0018] Step S21: Perform local minimum preselection on the original signal sequence of otoacoustic emission fine structure at the stimulation frequency, defining the frequency sequence as... The corresponding amplitude is This will satisfy the local minimum condition. All candidate points constitute the initial candidate minimum point set. ;

[0019] Step S22: For the candidate minimum point set Perform primary feature point screening, setting a primary threshold T1 and a neighbor range N1; for the point set If each local minimum point in the matrix passes the magnitude gradient constraint judgment rule on both its left and right sides, it is classified as a first-level feature point, forming a set of first-level feature points A.

[0020] Step S23: Perform secondary feature point screening on candidate points that fail the first-level screening. Set the secondary threshold T2 and the neighbor range N2. Sort the candidate points that fail the first-level screening in ascending order of amplitude. Select the first m points with the lowest energy for secondary screening. If both the left and right sides pass the amplitude gradient constraint judgment rule containing the third-level judgment rule, they are classified as secondary feature points, forming a secondary feature point set B.

[0021] Step S24: Set the feature maximum point screening threshold T3 and index range M, identify all local maxima in the original signal sequence of otoacoustic emission fine structure at the stimulus frequency, and classify the maxima points on both the left and right sides of the point set that pass the amplitude gradient constraint judgment rule as feature maximum points.

[0022] Step S25: Set the weight control point screening threshold T4. For ordinary minimum points, find the nearest feature maximum point within the index span M. If the magnitude difference between the minimum point and the nearest feature maximum point is greater than T4, it is classified as a weight control point, forming a weight control point set C.

[0023] Preferably, the first-level threshold T1 in step S22 is an empirical value, which is adaptively adjusted according to the signal-to-noise ratio characteristics of the test equipment; the amplitude gradient constraint judgment rule includes two levels: Level 1 rule determines whether the amplitude gradients on both sides of the minimum point are greater than T1; Level 2 rule, when Level 1 is not passed, determines whether at least one side has a gradient greater than T1 and the other side has a gradient greater than a preset secondary threshold.

[0024] Preferably, in step S23, the secondary threshold T2 is slightly lower than T1;

[0025] The magnitude gradient constraint determination rule includes three levels of determination rules, with an additional Level 3 rule: when Level 1 fails and Level 2 fails because the gradient on one side is insufficient, it is determined whether there is an adjacent compensation gradient on that side.

[0026] Preferably, in step 24, the feature maximum point screening threshold T3 is relatively small, and the requirement for the sharpness of the maximum point is low.

[0027] Preferably, in step 25, the weight control point screening threshold T4 is related to the expected volatility.

[0028] Preferably, step S3 specifically includes:

[0029] Step S31: Normalize the amplitudes of all feature points in set A and set B using inverse Min-Max mapping to obtain the normalized weights of the otoacoustic emission amplitudes at the stimulation frequency.

[0030] Step S32: The pure tone audiometry hearing threshold data is preprocessed using a piecewise linear mapping function, and the corresponding hearing threshold is matched with the otoacoustic emission feature point frequency for each stimulus frequency to obtain the hearing threshold mapping weight;

[0031] Step S33: Assign differentiated weight coefficients based on the feature point screening level. The magnitude factor weight coefficient for first-level feature points is: The magnitude factor weighting coefficient of the second-order feature points is The general weighting coefficient of the hearing threshold factor is γ, and the composite weight of the i-th feature point is denoted as γ. Calculate the composite weight for each feature point separately;

[0032] Step S34: Based on the frequency of all feature points and its composite weight Calculate the weighted average prediction frequency This is a preliminary prediction.

[0033] Preferably, in step S33:

[0034] For first-level feature points: ;

[0035] For secondary feature points: ;

[0036] in For the normalized weights of SFOAE amplitude, Weights are mapped to hearing thresholds; > .

[0037] Preferably, step S4 specifically includes:

[0038] Step S41: Divide the analysis frequency band into K consecutive logarithmic scale frequency bands according to the frequency interval of the otoacoustic emission test based on the stimulation frequency;

[0039] Step S42: For the set of weighted control points C, traverse each frequency band. If the number of weighted control points contained in a frequency band exceeds the preset threshold n, then the frequency band is determined to be a region with significant fluctuations.

[0040] Step S43: For each point within the identified significant fluctuation region, calculate the amplitude weight and distance weight respectively; where the amplitude weight adopts the reverse Min-Max mapping logic consistent with step b1, and the distance weight adopts the initial predicted frequency. A Gaussian function centered at the center;

[0041] Step S44: Combine the amplitude weight and distance weight to obtain the composite weight, and calculate the frequency offset. ;

[0042] Step S45: Adjust the frequency offset Compared with the initial predicted frequency Add them together to obtain the second-corrected frequency. .

[0043] Preferably, step S5 specifically includes:

[0044] Step S51: From all valid first- and second-level feature point sets, set a dual fusion criterion, including an index proximity criterion and a frequency proximity criterion, to locate the frequency relative to the preliminary predicted frequency. Or secondary correction frequency The fusion operation is triggered by the set of points that are closest in terms of index or frequency distance.

[0045] Step S52: Based on whether there is a valid secondary correction result, select the corresponding fusion path for weighted fusion calculation to obtain the fusion prediction frequency. .

[0046] Preferably, in step S6:

[0047] If a second correction is performed and conditional fusion is triggered, then = ;

[0048] If a second correction is made but conditional fusion is not triggered, then = ;

[0049] If a second correction is not performed but conditional fusion is triggered, then = ;

[0050] If neither secondary correction nor conditional fusion is triggered, then = .

[0051] This invention provides an objective method for detecting tinnitus pitch based on the fine structure of otoacoustic emissions at stimulation frequency. It has the following beneficial effects:

[0052] By collecting SFOAE fine structure signals and PTA hearing threshold data from the tinnitus ear of the subject, a multi-level threshold screening strategy is used to accurately extract primary feature valley points, secondary feature valley points and weight control points from the SFOAE fine structure signals, effectively identifying feature signals related to tinnitus frequency. Objective prediction of tinnitus frequency can be completed without active feedback from the patient, overcoming the dependence of subjective fitting methods on patient cooperation.

[0053] The data transfer relationships between each step are clearly defined: the SFOAE fine structure signal acquired in step S1 is input into step S2 for feature extraction; the set of primary and secondary feature valley points extracted in step S2 is input into step S3 for composite weighted prediction; the set of weight adjustment points extracted in step S2 is input into step S4 for secondary correction; the set of primary and secondary feature valley points extracted in step S2 is input into step S5 for nearest feature point conditional fusion; and step S6 determines the final output based on the process path. The data sources for each step are clear, avoiding data gaps or confusion.

[0054] By integrating SFOAE amplitude normalization weights and hearing threshold mapping weights for two-factor composite weighted prediction, the complementary information of local cochlear functional state and overall auditory sensitivity is fully utilized, overcoming the limitations of single-factor prediction and significantly improving the reliability and accuracy of prediction results.

[0055] By employing a significant fluctuation region identification and secondary correction mechanism, dense functional fluctuations of the SFOAE fine structure signal within a narrow frequency range are captured, allowing for the calibration of preliminary global predictions based on physiological details. Simultaneously, a conditional fusion strategy based on nearest feature points effectively integrates the model's computational inferences with direct observational evidence from the data, further enhancing the stability and accuracy of the final prediction.

[0056] This invention solves many problems faced by subjective fitting, especially providing an objective testing method for special groups such as those with cognitive impairment who are unable to make accurate subjective feedback. It provides a reliable data foundation for the development of personalized sound therapy plans and has good clinical application value and promotion prospects. Attached Figure Description

[0057] Figure 1 This is a flowchart of the overall method for objectively predicting tinnitus frequency based on stimulus-frequency otoacoustic emissions (SFOAE) fine-structure signals and pure-tone audiometry (PTA).

[0058] Figure 2 A flowchart for multi-level threshold extraction of fine-structure feature signals from stimulation frequency otoacoustic emissions (SFOAE).

[0059] Figure 3 A schematic diagram of the stimulus frequency otoacoustic emission (SFOAE) for predicting the results (including feature point extraction and prediction points at each step). Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Example 1

[0062] The following is in conjunction with the appendix Figures 1 to 3 The complete implementation process of the present invention will be described in detail.

[0063] Figure 1 The flowchart shows the overall process framework of the objective prediction method for tinnitus frequency based on SFOAE fine structure signal and PTA, illustrating the data transfer relationship between steps S1 to S6.

[0064] Figure 2 The flowchart for multi-level threshold extraction of SFOAE fine structure feature signals shows the detailed process of initial candidate minimum point set pre-selection, first-level feature point screening, second-level feature point screening, feature maximum point determination and weight adjustment point determination in step S2.

[0065] Figure 3 This is a schematic diagram of the predicted SFOAE, showing the original signal of the SFOAE fine structure, the feature point extraction results at each level, and the prediction points at each step. The position of ) in the spectrum.

[0066] In one specific embodiment, the present invention provides an objective detection method for tinnitus pitch based on the fine structure of otoacoustic emissions at stimulation frequency, the specific implementation process of which is as follows:

[0067] I. Subject Screening and Preparation

[0068] Subjective tinnitus patients with a disease course of no less than 6 months were selected. Subjects were required to have no organic lesions in the external auditory canal and middle ear (confirmed by otoscopy and acoustic impedance examination), possess normal cognitive and behavioral abilities, and be able to cooperate in completing all test procedures.

[0069] II. SFOAE Fine Structure Signal Acquisition

[0070] The test was conducted using an OAE detection device based on an exponential sweep frequency mode.

[0071] The lower limit frequency of the sweep frequency is Set to 125Hz, with the upper frequency limit being... The frequency was set to 12000Hz, covering the common frequency range of tinnitus. During the test, the frequency ratio of the suppressor to the probe tone was strictly controlled to a fixed value (preferably 1.2).

[0072] The probe tone intensity is set within a reasonable test range (preferably 60 dB SPL), and tests are performed according to a set test step size (preferably logarithmic intervals, with at least 10 test points per octave). The suppressor tone intensity maintains a fixed dynamic relationship with the probe tone (preferably the suppressor tone intensity is 10 dB higher than the probe tone). The duration of a single frequency sweep is set to... (Preferred time is 2 seconds), and the data is averaged n times during the acquisition process (preferred n≥30).

[0073] This effectively suppresses environmental and physiological background noise and improves the signal-to-noise ratio. Through the above parameter configuration, fine structure signals of otoacoustic emissions (SFOAE) at stimulation frequencies covering the range of 125Hz to 12000Hz are accurately induced and extracted.

[0074] III. PTA Hearing Threshold Data Acquisition

[0075] Air conduction pure-tone audiometry was used to quantitatively assess the binaural hearing sensitivity of the subjects. During the test, the subjects wore over-ear headphones and responded to pure-tone stimuli of different sound pressure levels output by the headphones in a soundproof environment (background noise ≤30dBA). Hearing thresholds were determined using a ramp-and-drop method (5dB ramp-and-drop, 10dB drop), and the air conduction hearing thresholds of the subjects were recorded at N standard frequency points from 125Hz to 16000Hz (N≥2, preferably N=11, standard frequency points are 125, 250, 500, 750, 1000, 1500, 2000, 3000, 4000, 6000, 8000, 10000, 12000, 16000Hz).

[0076] IV. Multi-level threshold feature extraction of SFOAE fine-structure signals

[0077] See attached document Figure 2 The process shown involves multi-level threshold feature extraction of the SFOAE fine structure signal acquired in step S1:

[0078] Step S21: Local Minimum Pre-selection. Local minima are identified in the original signal sequence of the fine structure of stimulus-frequency otoacoustic emissions (SFOAE), and the frequency sequence is defined as... The corresponding amplitude is By traversing all data points, the condition for a local minimum will be satisfied. All candidate points constitute the initial candidate minimum point set P.

[0079] Step S22: Primary Feature Point Screening. Set a primary threshold T1 (preferably 12dB), and the neighbor range N1 as the preset search depth. For each minimum point in the point set P, a two-stage judgment rule is adopted: Level 1 rule determines whether the gradient magnitudes on both sides of the minimum point are greater than T1; if so, it passes. Level 2 rule, if Level 1 fails, further determines whether at least one side has a gradient greater than T1 and the other side has a gradient greater than the preset secondary threshold; if so, it passes. Both sides of the minimum point must pass the judgment rule simultaneously to be classified as a primary feature point, forming the primary feature point set A. Primary feature points undergo more rigorous screening and have higher confidence.

[0080] Step S23: Secondary Feature Point Screening. Set a secondary threshold T2 (preferably 10dB), and the neighbor range N2 is the preset search depth. For points in point set P that fail the primary screening, sort their amplitudes in ascending order, and select the m points with the lowest energy (preferably m=8) for secondary screening to avoid missing important feature points. The judgment rules include three levels: a Level 3 rule is added, which determines whether there is an adjacent compensating gradient on that side (i.e., whether there is a significant increase in gradient within a more distant neighbor range) when Level 1 fails and Level 2 fails due to insufficient gradient on one side. The left and right sides of the minimum point must pass the judgment rules simultaneously to be classified as secondary feature points, forming the secondary feature point set B.

[0081] Step S24: Characteristic Maximal Point Determination. Set a threshold T3 (preferably 2dB) and an index range M. Identify all local maxima in the original signal sequence of the fine structure of stimulus frequency otoacoustic emission (SFOAE). Maximal points that pass the amplitude gradient constraint determination rule on both the left and right sides of the point set are classified as characteristic maxima. The characteristic maxima selection threshold T3 is relatively small, and the requirement for sharpness of the maxima is low.

[0082] Step S25: Determining Weighted Control Points. Set a threshold T4 (related to the expected volatility, preferably 5 dB). For ordinary minimum points (i.e., minimum points not classified as primary or secondary feature points), search for the nearest eigenmaxima within the index span M. If the amplitude difference between this minimum point and the nearest eigenmaxima is greater than T4, it is classified as a weighted control point, forming a weighted control point set C. The weighted control point screening threshold T4 is related to the expected volatility. Weighted control points reflect local strong contrast energy dips in the stimulus frequency otoacoustic emission (SFOAE) signal.

[0083] V. SFOAE-PTA Two-Factor Composite Weighted Calculation

[0084] Based on the set of primary feature valley points A and the set of secondary feature valley points B extracted in step S2, the following calculations are performed:

[0085] Step S31: SFOAE amplitude normalization and weight calculation; The amplitudes of all feature points in sets A and B are normalized using an inverse Min-Max mapping. The inverse Min-Max mapping formula considers the extreme values ​​of the amplitudes of all feature points for the current patient and introduces a small offset to prevent weights from being zero. The maximum value of all feature points for the current patient. and minimum value Calculate the weights; points with smaller amplitudes receive larger normalized values.

[0086] Step S32: Hearing threshold mapping and weight calculation; PTA hearing threshold data are preprocessed using a piecewise linear mapping function. The piecewise linear mapping function eliminates the influence of negative hearing threshold values ​​while preserving hearing threshold differences.

[0087] To integrate individualized hearing information, a corresponding hearing threshold needs to be assigned to each SFOAE feature point frequency. Since clinical PTA is performed only at discrete standard frequency points, and given that human auditory frequency perception follows an approximately logarithmic law, the logarithmic frequency nearest neighbor matching principle is used for approximation, extracting the corresponding hearing threshold from the patient's hearing threshold vector. The mapped and approximated hearing thresholds are then normalized to obtain the hearing threshold mapping weights. .

[0088] Step S33: Calculate the SFOAE-PTA two-factor composite weight; assign differentiated weight coefficients based on the feature point screening hierarchy. The composite weight is denoted as... The composite weight of the i-th feature point is denoted as . Composite weight The calculation formula is:

[0089] For first-level feature points: ;

[0090] For secondary feature points: ;

[0091] in The preferred value is 0.5. The preferred value is 0.3. The preferred value is 0.2; > .

[0092] Step S34: Calculate the weighted average prediction frequency; based on the frequency of all feature points. and its composite weight According to the formula Calculate the weighted average prediction frequency This is a preliminary prediction.

[0093] VI. Identification of Significant Fluctuation Regions and Secondary Correction

[0094] Based on the set of weight control points C extracted in step S2, the preliminary prediction results obtained in step S3 are analyzed. Make corrections:

[0095] Step S41: Divide the analysis frequency band into K consecutive logarithmic scale bands based on the frequency interval of the stimulus frequency otoacoustic emission (SFOAE) test.

[0096] Step S42: For the set C of weighted control points further screened from the first and second level screenings, traverse each frequency band. If the number of weighted control points contained in a frequency band exceeds the preset threshold n (preferably n=3), then the frequency band is determined to be a region with significant fluctuations, which aims to make subsequent corrections to regions that indicate large fluctuations in local functions.

[0097] Step S43: For all points within the identified significant fluctuation regions, calculate the amplitude weight and distance weight; the amplitude weight uses the same reverse Min-Max mapping logic as in step S31, and the distance weight uses a Gaussian weighting function.

[0098] ,

[0099] Based on preliminary frequency prediction Centered on, For bandwidth parameters (preferably 1000Hz).

[0100] Step S44: Combining the above two types of local features, obtain the composite weight for each point: ;

[0101] The frequency offset is calculated by weighted averaging of the frequency and the deviation at each point within the region. ;

[0102] Step S45: Predicted tinnitus frequency after secondary correction: The secondary correction is activated only when a region of significant fluctuation is detected.

[0103] The secondary correction, as a conditional execution branch in the algorithm flow, is activated only when a region of significant fluctuation is detected. By focusing on specific local frequency bands of the cochlear stimulation frequency otoacoustic emission (SFOAE) signal fluctuations and finely weighting the SFOAE characteristics of that region, a physiological detail-based calibration of the preliminary global prediction results is provided, thereby improving the prediction accuracy for complex cases.

[0104] VII. Conditional Fusion Based on Nearest Feature Points

[0105] Based on the nearest feature points selected from the set of primary feature valley points A and the set of secondary feature valley points B extracted in step S2, conditional fusion is performed:

[0106] Specifically, step S5 is based on a conditional fusion strategy using the nearest feature points. This is done after obtaining the initial prediction frequency. (or after two revisions) Following this, our algorithm introduces a nearest feature point conditional fusion strategy. This strategy aims to address a common situation where predicted values ​​generated by weighted averaging may sometimes deviate numerically from the patient's actual tinnitus frequency, while a salient feature point inherent in the stimulation frequency-of-otoacoustic emission (SFOAE) data may have a closer frequency. The core idea of ​​this strategy is that when the initial prediction is sufficiently close to the nearest feature point in frequency or data index, it indicates that they may be capturing different aspects of the same physiological source. Fusing them effectively integrates the computational inferences of the model with direct observational evidence from the data, thereby improving the stability and accuracy of the final prediction.

[0107] Step S51: Recent Feature Point Identification and Evaluation. Based on the preliminary prediction frequency... As anchor point (if a valid secondary correction result exists, then the secondary correction frequency shall be used). Using anchor points, all primary and secondary feature points are traversed to filter out feature points within the index range x (preferably x=8) or the frequency range yHz (preferably y=250). The index criterion is based on a consistency check of data sampling density to ensure that two frequency points are in adjacent regions in the original data space. If the frequency criterion is less than the equivalent rectangular bandwidth of the auditory filter in the relevant frequency band, it means that the two frequencies may be processed by the same auditory channel.

[0108] Step S52: Dual-path mean fusion calculation. Depending on whether a valid secondary correction result exists, the fusion is performed via two paths. The arithmetic mean of the current prediction result and the frequency values ​​of all selected feature points is taken to obtain the fused prediction frequency. .

[0109] 8. Final Predicted Frequency Output

[0110] Based on the entire prediction process described above, and depending on whether the data has undergone secondary correction and condition fusion, the prediction frequency obtained in the last step is set as the final prediction frequency. :

[0111] If a second correction is performed and conditional fusion is triggered, then = (based on (fusion results)

[0112] If a second correction is made but conditional fusion is not triggered, then = ;

[0113] If a second correction is not performed but conditional fusion is triggered, then = (based on (fusion results)

[0114] If neither secondary correction nor conditional fusion is triggered, then = .

[0115] Example 2

[0116] In another specific embodiment, the method of the present invention can be further optimized in terms of parameter settings to adapt to different clinical application scenarios:

[0117] For patients with high-frequency tinnitus (tinnitus frequency > 6000 Hz), the high-frequency feature point screening threshold can be lowered to improve the accuracy of high-frequency detection and obtain fine structural information with higher resolution in the high-frequency band.

[0118] For patients with low-frequency tinnitus (tinnitus frequency <500Hz), the low-frequency feature point screening threshold can be lowered to improve the accuracy of low-frequency detection, and the screening of low-frequency weight control points can be enhanced to ensure that more effective data of low-frequency SFOAE signals are included in the frequency calculation.

[0119] In addition, for special patient groups who cannot cooperate with long-term testing (such as children and patients with cognitive impairment), the frequency range of SFOAE fine structure sweep or the number of test points can be shortened to keep the test time within an acceptable range (preferably ≤15 minutes) while ensuring prediction accuracy.

[0120] Example 3

[0121] In yet another specific embodiment, the method of the present invention can be combined with artificial intelligence technology to achieve automated and intelligent tinnitus frequency prediction:

[0122] The entire signal processing and feature extraction process described in steps S1 to S6 is integrated into an automated analysis software module and embedded into the data processing system of the OAE detection device. The software module has a built-in parameter system (including optimal values ​​for key parameters such as T1, T2, T3, T4, α, β, γ, and σ) trained and optimized with a large number of clinical samples. It can automatically complete the entire calculation process based on the input patient SFOAE and PTA data and output the final predicted frequency. And the corresponding confidence level assessment results.

[0123] The software module also provides a visualization interface to display the original signal of SFOAE fine structure, the marked positions of feature points at each level, and the prediction results at each step. , , The system includes indicators for areas of significant fluctuation, which facilitates clinicians in visually reviewing and confirming the prediction results.

[0124] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An objective method for detecting tinnitus pitch based on the fine structure of otoacoustic emissions at stimulation frequency, characterized in that, include: Step S1: Collect the stimulation frequency otoacoustic emission fine structure signal and pure tone audiometry threshold data of the subject's tinnitus ear; Step S2: Perform multi-level threshold feature extraction on the otoacoustic emission fine structure signal at the stimulation frequency to obtain the set of first-level feature valley points, the set of second-level feature valley points, and the set of weighted adjustment points; Step S3: Based on the set of first-level feature valley points and the set of second-level feature valley points, calculate the normalized weight of the stimulation frequency otoacoustic emission amplitude and the mapping weight of the hearing threshold respectively, and perform a two-factor composite weighted calculation of stimulation frequency otoacoustic emission-pure tone audiometry to obtain the preliminary predicted frequency; Step S4: Based on the set of weight adjustment points, identify and correct the significant fluctuation areas of the obtained preliminary prediction results. Step S5: Based on the nearest feature points selected from the extracted first-level feature valley point set and second-level feature valley point set, perform conditional fusion optimization on the prediction results to obtain the final prediction frequency; Step S6: Based on the entire prediction process described above, and depending on whether the data has undergone secondary correction and condition fusion, set the prediction frequency obtained in the last step as the final prediction frequency. .

2. The method according to claim 1, characterized in that, In step S1: The acquisition process of the stimulation frequency otoacoustic emission fine structure signal is as follows: The stimulation signal is acquired using an exponential sweep frequency mode, with the lower sweep frequency being... The upper limit frequency is During testing, the frequency ratio of the suppressor tone to the probe tone was strictly controlled to a fixed value, the probe tone intensity was set within a reasonable testing range, and testing was conducted according to the set testing step size. The suppressor tone intensity and the probe tone maintained a fixed dynamic relationship, and the duration of a single frequency sweep was set to [missing value]. And the data is averaged N times during the collection process; The process of collecting pure-tone audiometry threshold data is as follows: using air conduction pure-tone audiometry, the hearing threshold is determined using the rise and fall method, and the air conduction hearing threshold of the subject is recorded at N standard frequency points from 125Hz to 16000Hz.

3. The method according to claim 1, characterized in that, Step S2 specifically includes: Step S21: Perform local minimum preselection on the original signal sequence of otoacoustic emission fine structure at the stimulation frequency, defining the frequency sequence as... The corresponding amplitude is All candidate points that satisfy the local minimum condition are used to form an initial candidate minimum point set. ; Step S22: For the candidate minimum point set Perform primary feature point screening, setting a primary threshold T1 and a neighbor range N1; for the point set If each local minimum point in the matrix passes the magnitude gradient constraint judgment rule on both its left and right sides, it is classified as a first-level feature point, forming a set of first-level feature points A. Step S23: Perform secondary feature point screening on candidate points that fail the first-level screening. Set the secondary threshold T2 and the neighbor range N2. Sort the candidate points that fail the first-level screening in ascending order of amplitude. Select the first m points with the lowest energy for secondary screening. If both the left and right sides pass the amplitude gradient constraint judgment rule containing the third-level judgment rule, they are classified as secondary feature points, forming a secondary feature point set B. Step S24: Set the feature maximum point screening threshold T3 and index range M, identify all local maxima in the original signal sequence of otoacoustic emission fine structure at the stimulus frequency, and classify the maxima points on both the left and right sides of the point set that pass the amplitude gradient constraint judgment rule as feature maximum points. Step S25: Set the weight control point screening threshold T4. For ordinary minimum points, find the nearest feature maximum point within the index span M. If the magnitude difference between the minimum point and the nearest feature maximum point is greater than T4, it is classified as a weight control point, forming a weight control point set C.

4. The method according to claim 3, characterized in that, In step S22: The first-level threshold T1 is an empirical value, which is adaptively adjusted according to the signal-to-noise ratio characteristics of the test equipment. The magnitude gradient constraint determination rule includes two levels: Level 1 rule: Determine whether the magnitude gradients on both sides of the minimum point are greater than T1; Level 2 rule: If Level 1 is not passed, determine whether there is at least one gradient greater than T1 and the other gradient greater than a preset secondary threshold.

5. The method according to claim 3, characterized in that, In step S23, the secondary threshold T2 is slightly lower than T1; The magnitude gradient constraint determination rule includes three levels of determination rules, with an additional Level 3 rule: when Level 1 fails and Level 2 fails because the gradient on one side is insufficient, it is determined whether there is an adjacent compensation gradient on that side.

6. The method according to claim 1, characterized in that, Step S3 specifically includes: Step S31: Normalize the amplitudes of all feature points in set A and set B using inverse Min-Max mapping to obtain the normalized weights of the otoacoustic emission amplitudes at the stimulation frequency. Step S32: The pure tone audiometry hearing threshold data is preprocessed using a piecewise linear mapping function, and the corresponding hearing threshold is matched with the otoacoustic emission feature point frequency for each stimulus frequency to obtain the hearing threshold mapping weight; Step S33: Assign differentiated weight coefficients according to the feature point screening level. The magnitude factor weight coefficient for first-level feature points is: The magnitude factor weighting coefficient of the second-order feature points is The general weighting coefficient of the hearing threshold factor is γ, and the composite weight of the i-th feature point is denoted as γ. Calculate the composite weight for each feature point separately; Step S34: Based on the frequency of all feature points and its composite weight Calculate the weighted average prediction frequency This is a preliminary prediction.

7. The method according to claim 6, characterized in that, In step S33: For first-level feature points: ; For secondary feature points: ; in For the normalized weights of SFOAE amplitude, Weights are mapped to hearing thresholds; > .

8. The method according to claim 1, characterized in that, Step S4 specifically includes: Step S41: Divide the analysis frequency band into K consecutive logarithmic scale frequency bands according to the frequency interval of the otoacoustic emission test based on the stimulation frequency; Step S42: For the set of weighted adjustment points C, traverse each frequency band. If the number of weighted adjustment points contained in a frequency band exceeds the preset threshold n, then the frequency band is determined to be a region with significant fluctuations. Step S43: For each point within the identified significant fluctuation region, calculate the amplitude weight and distance weight respectively; where the amplitude weight uses the inverse Min-Max mapping logic, and the distance weight uses the initial predicted frequency. A Gaussian function centered at the center; Step S44: Combine the amplitude weight and distance weight to obtain the composite weight, and calculate the frequency offset. ; Step S45: Adjust the frequency offset Compared with the initial predicted frequency Add them together to obtain the second-corrected frequency. .

9. The method according to claim 1, characterized in that, Step S5 specifically includes: Step S51: From all valid first- and second-level feature point sets, set a dual fusion criterion, including an index proximity criterion and a frequency proximity criterion, to locate the frequency relative to the preliminary predicted frequency. Or secondary correction frequency The fusion operation is triggered by the set of points that are closest in terms of index or frequency distance. Step S52: Based on whether there is a valid secondary correction result, select the corresponding fusion path for weighted fusion calculation to obtain the fusion prediction frequency. .

10. The method according to claim 1, characterized in that, In step S6: If a second correction is performed and conditional fusion is triggered, then = ; If a second correction is made but conditional fusion is not triggered, then = ; If a second correction is not performed but conditional fusion is triggered, then = ; If neither secondary correction nor conditional fusion is triggered, then = .

Images