High-definition imaging method for cochlear implant bone groove area and endoscopic imaging cochlea based on cochlear implant

Abstract

This invention relates to the field of image processing technology, and particularly to a high-definition imaging method for the bony alveolar region of a cochlear implant and endoscopic imaging of the cochlea. The invention proposes the following scheme: the method identifies bony areas with strong reflections based on raw imaging data, constructs a bone reflection source map, and extracts local geometric features such as the normal direction and curvature of the region; based on this, it determines the relative distance information of the location to be processed relative to the bony areas with strong reflections, and constructs a bone reflection contamination propagation kernel associated with the target boundary point by combining the acquired parameters; by generating a contamination contribution field and applying normal consistency constraints, it performs restricted digital hidden surface compensation on the affected area. This application can suppress directional diffusion contamination caused by strong reflections while preserving the real tissue texture, improving the clarity and stability of imaging the bony alveolar region of the cochlear implant.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a high-definition imaging method for the bone groove region of an implant based on a cochlear implant and endoscopic imaging of the cochlea. Background Technology

[0002] During endoscopic imaging of the cochlear cavity, the cavity walls are often highly reflective, and their spatial structure is curved or spiral, causing strong reflections of the imaging light at local interfaces. These strong reflections not only create bright, saturated areas at the interfaces but also diffuse into neighboring areas in a certain direction, causing abnormal grayscale gradations and compression of detail contrast, thus interfering with the true texture of the deep alveolar region. Especially under near-field imaging conditions, the limited distance between the probe and the cavity wall means that changes in acquisition parameters such as gain, exposure, and scanning angle further amplify the impact of strong reflections on the neighborhood, resulting in contamination features in the image that resemble the actual structural morphology. Existing techniques often use threshold filtering, local smoothing, or overall contrast adjustment to process bright areas, but these methods typically only suppress the strong reflection areas themselves, failing to address their directional diffusion to the outer neighborhoods. This can easily lead to weakened details in the true alveolar region or unsuppressed residual contamination. Therefore, targeted control of the impact of strong reflections on the neighborhood, while ensuring the integrity of the cochlear alveolar region's structural information, has become a pressing technical problem.

[0003] To address the above issues, this application presents a high-definition imaging method for the bone groove region of the cochlear implant and an endoscopic imaging method for the cochlea. Summary of the Invention

[0004] The technical problem to be solved by this invention is to address the shortcomings of existing technologies by providing a high-definition imaging method for the bone groove region of an implanted cochlear implant and an endoscopic imaging method for the cochlear implant. The method identifies bony strong reflection regions based on raw imaging data, constructs a bone reflection source map, and extracts local geometric features such as the normal direction and curvature of the region. Based on this, it determines the relative distance information of the location to be processed relative to the bony strong reflection region, and constructs a bone reflection contamination propagation kernel associated with the target boundary point by combining the acquired parameters. By generating a contamination contribution field and applying normal consistency constraints, it performs restricted digital blanking compensation on the affected area.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A high-resolution imaging method for the alveolar region of a cochlear implant, the method comprising: Acquire raw imaging data of the alveolar region and the acquisition parameter information corresponding to the raw imaging data; Based on the original imaging data, the bony strong reflective regions in the alveolar region are identified to obtain a bone reflective source map, wherein the bone reflective source map at least characterizes the spatial distribution of the bony strong reflective regions and the boundary information of the bony strong reflective regions. Based on the bone reflex source map, local geometric features of the bony strong reflex region are extracted, and the local geometric features include at least the region normal direction and the region curvature; Based on the local geometric features, the relative distance information of the position to be processed in the original imaging data with respect to the bony strong reflective area is determined. The bone reflection contamination propagation kernel is constructed in combination with the acquired parameter information. The position to be processed represents the pixel position in the original imaging data that needs to be evaluated for bone reflection contamination. Based on the bone reflection source map and the bone reflection contamination propagation nucleus, digital blanking compensation is performed on the original imaging data to obtain the compensated cochlear alveolar region imaging result.

[0006] The identification methods based on the aforementioned bony strong reflex areas include: The original imaging data is preprocessed to obtain preprocessed imaging data. The preprocessing includes at least one of noise reduction processing and grayscale normalization processing. High-brightness candidate regions are extracted based on the preprocessed imaging data to obtain strong bony reflection candidate regions. Boundary detection is then performed on the strong bony reflection candidate regions to extract the candidate region boundary information. Based on the connectivity features of the candidate regions for strong bony reflexes and the boundary information of the candidate regions, the candidate regions for strong bony reflexes are screened to determine the strong bony reflex regions.

[0007] The methods for extracting the local geometric features include: Boundary point sets are extracted from the bony strong reflection regions in the bone reflection source map to obtain the boundary point sequence of the bony strong reflection regions; Select a target boundary point from the boundary point sequence, and determine a set of boundary neighborhood points with the target boundary point as the center of a preset neighborhood range; Based on the set of boundary neighborhood points, curve fitting is performed on the local boundary of the bony strong reflex region to obtain the tangential direction at the target boundary point, and the regional normal direction at the target boundary point is determined according to the tangential direction. The local curvature at the target boundary point is determined based on the curve fitting results, wherein the local curvature is characterized by the reciprocal of the radius of curvature formed by the set of boundary neighborhood points. The region's normal direction and curvature are summed at multiple target boundary points of the bony strong reflective region to obtain the local geometric features of the bony strong reflective region.

[0008] The target boundary points are selected as follows: the boundary point sequence is sampled at equal intervals according to a preset sampling interval to obtain an initial sampled boundary point set; for each initial sampled boundary point in the initial sampled boundary point set, the boundary reflection intensity value of the initial sampled boundary point is extracted from the original imaging data, the boundary reflection intensity value is determined at least by the pixel grayscale value at the initial sampled boundary point and the proportion of bright pixels in its neighborhood; the initial sampled boundary point set is subjected to intensity-weighted filtering based on the boundary reflection intensity value, so as to select the initial sampled boundary point whose boundary reflection intensity value meets the preset intensity condition as the target boundary point, and the boundary reflection intensity value is used as the kernel intensity parameter of the corresponding target boundary point.

[0009] The methods for determining the relative distance information include: For each target boundary point in the bony strong reflective region, the outer normal direction is determined based on the regional normal direction at the target boundary point, and a discrete sampling chain is generated along the outer normal direction starting from the target boundary point. The discrete sampling chain consists of multiple candidate pixel positions obtained by expanding outward point by point according to a preset sampling step size. In the discrete sampling chain, at least one target pixel position corresponding to the position to be processed is determined based on a preset point selection rule. The preset point selection rule includes at least the inflection point determination of the grayscale attenuation index, the peak value determination of the gradient magnitude under the condition that the gradient direction points to the target boundary point, and the valley value determination of the local contrast compression index. Establish a mapping relationship between the target pixel position and the corresponding target boundary point, and use the corresponding target boundary point as the distance reference point of the target pixel position; Based on the projection interval between the target pixel position and the corresponding target boundary point in the outer normal direction, the relative distance information of the target pixel position relative to the bony strong reflective region is determined, wherein the projection interval is characterized by the preset sampling step size and the index of the target pixel position in the discrete sampling chain.

[0010] A bone reflex contamination transmission kernel is constructed by combining the collected parameter information, including: Based on the acquired parameter information, a set of nuclear modulation parameters is extracted to characterize the degree of strong reflection response expansion. The set of nuclear modulation parameters includes gain parameters, exposure parameters, emission energy parameters, and scanning angle parameters. Based on the local geometric features, the directional morphological parameters and main diffusion direction of the bone reflection pollution propagation nucleus are determined, wherein the directional morphological parameters include diffusion scale parameters along the normal direction of the region and diffusion scale parameters along the tangential direction of the region. Based on the relative distance information, the distance attenuation parameter of the bone reflection contamination propagation core is determined; The directional morphological parameters and the distance attenuation parameters are adjusted according to the nuclear modulation parameter set to generate a bone reflection contamination propagation nucleus.

[0011] Adjusting the directional shape parameter and the distance attenuation parameter includes: A kernel intensity adjustment coefficient is determined based on the gain parameter and the exposure parameter, and the diffusion scale parameter along the normal direction of the region and the diffusion scale parameter along the tangential direction of the region are synchronously amplified or reduced according to the kernel intensity adjustment coefficient. A directional offset angle is determined based on the scanning angle parameters, and the main diffusion direction of the bone reflection contamination propagation nucleus is rotated and corrected according to the directional offset angle. An attenuation adjustment coefficient is determined based on the emitted energy parameters, and the distance attenuation parameters are adjusted according to the attenuation adjustment coefficient. Bone reflection contamination propagation nuclei are generated based on the adjusted diffusion scale parameters along the regional normal direction, the adjusted diffusion scale parameters along the regional tangential direction, the adjusted main diffusion direction, and the adjusted distance attenuation parameters.

[0012] The bone reflection pollution propagation kernel is characterized by a set of kernel parameters that are associated one-to-one with each target boundary point of the strong bone reflection region. The set of kernel parameters includes kernel intensity parameters, diffusion scale parameters along the normal direction, diffusion scale parameters along the tangential direction, main diffusion direction parameters, and distance attenuation parameters. Based on the set of kernel parameters, a kernel weight function is determined in a local coordinate system with the corresponding target boundary point as the origin and formed by its tangential and normal directions.

[0013] The compensation methods for the digital blanking compensation include: Based on the bone reflectance source map and the bone reflectance contamination propagation kernel, a contamination contribution field of the outer neighborhood is generated for each target boundary point of the strong bony reflectance region in the corresponding local coordinate system to obtain the upper bound information of the contamination contribution of each outer neighborhood position. For the location to be processed, the limited compensation amount for the location to be processed is determined based on the relative distance information relative to the corresponding target boundary point and the pollution contribution upper bound information. A normal sampling chain consistency constraint is constructed based on the outer normal direction associated with the target boundary point, and the constrained compensation amount is corrected for normal consistency based on the normal sampling chain consistency constraint. The corresponding unprocessed position is compensated based on the limited compensation amount after normal consistency correction.

[0014] Endoscopic imaging cochlea, the endoscopic imaging cochlea comprising: A cochlear mold, comprising an insertion part and a probe end disposed at the distal end of the insertion part, the probe end being used to enter the cochlear cavity and perform near-field imaging of the cochlear alveolar region; An imaging component is disposed at the end of the probe. The imaging component includes an imaging sensor and an imaging optics corresponding to the optical axis of the imaging sensor, and is used to acquire raw imaging data of the alveolar region. An illumination component, disposed at the end of the probe and facing the framing direction, is used to provide illumination light to the cochlear alveolar region to support imaging by the imaging component; A processing component, electrically connected to the imaging component, is configured to generate a bone reflection source map based on the original imaging data, extract local geometric features of the bony strong reflection area, determine the relative distance information of the location to be processed relative to the bony strong reflection area, construct a bone reflection contamination propagation kernel by combining the acquired parameter information, and perform digital blanking compensation on the original imaging data based on the bone reflection source map and the bone reflection contamination propagation kernel to output the cochlear alveolar region imaging result; An output component, electrically connected to the processing component, is used to output the compensated cochlear alveolar bone region imaging results to a display terminal.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention constructs a bone reflection source map and extracts local geometric features of areas with strong bone reflection. Based on clearly defined regional normal direction and curvature information, it introduces relative distance modeling of the location to be processed relative to the strong reflection interface. Combined with acquired parameters, it generates bone reflection contamination propagation kernels that are one-to-one associated with target boundary points. This allows the impact of strong reflections on the neighborhood to be characterized in a directional and distance-dependent manner. It can spatially define the compensation range and achieve progressive correction under normal consistency constraints, thereby avoiding the loss of detail caused by over-smoothing. Attached Figure Description

[0016] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 An exemplary application scenario diagram provided for an embodiment of this application; Figure 2 This is a schematic diagram of the structure of the endoscopic imaging cochlea provided in an embodiment of this application; Figure 3 This is a flowchart illustrating a high-definition imaging method for the bone groove region of a cochlear implant based on an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0019] The high-definition imaging method for the cochlear implant alveolar region proposed in this application addresses a typical endoscopic near-field imaging scenario in clinical and research settings: the cochlea is surrounded by dense bony structures with strong interface reflections, and the alveolar region is often located in a curved and narrow deep field of view. The imaging process is simultaneously affected by factors such as limited viewing angle, limited illumination path, and tissue scattering attenuation. Especially in actual operation, the available space for posture adjustment at the probe tip is limited, and acquisition is often only possible from one side or a few angles. When bony walls, bony protrusions, or highly reflective interfaces appear in the field of view, the strong reflective signals will create significant local highlights and neighborhood grayscale increases in the imaging chain, which in turn will dynamically crowd out the weakly reflective alveolar region near the interface. This causes the subtle textures, boundary transitions, and weak contrast structures of the cochlear alveolar region to be submerged or covered by trailing artifacts, resulting in unstable results such as overall clarity but indistinguishable alveolar regions or bright bony walls but lack of detail.

[0020] In current technological practices, the common approach is to pursue an overall visual appeal through automatic exposure, automatic gain, or global contrast enhancement, supplemented by local highlight reduction, glare reduction, or simple filtering to mitigate the overexposure effects in areas of strong reflection. However, for scenarios like the cochlea where strong contrast, strong reflection, and weak signals coexist, the above strategies are often difficult to achieve simultaneously: global parameter adjustments are easily influenced by bony areas with strong reflections, resulting in the grayscale of the alveolar region being compressed into a low dynamic range; while suppressing only the bright areas often leads to the accidental clipping of neighboring textures and the destruction of the true shape of the boundaries, and even inconsistencies in compensation under different acquisition postures and lighting conditions, resulting in insufficient reproducibility and interpretability of alveolar region details.

[0021] It is important to emphasize that strong reflection does not only cause local saturation points. Its impact on the neighborhood often presents as directional diffusion and gradual contamination. Moreover, this contamination is coupled with changes in acquisition parameters such as the geometry of the bony interface, the scanning angle of the probe, and the exposure gain, making it difficult to handle stably with a single threshold or a single filtering strategy.

[0022] It should be noted that the contamination in this application is not the random degradation caused by random noise, speckle noise, or sensor thermal noise in the traditional sense, nor is it the same as the visible bright area caused by overexposure alone. The contamination described in this application focuses more on a signal carrying capacity encroachment phenomenon triggered by the bony strong reflective interface. Its essence lies in the fact that after the strong reflective response is superimposed on the local tissue propagation conditions through the imaging link, it introduces non-target intensity contributions to the outer neighborhood, which squeezes out the grayscale space in the neighborhood that was originally used to characterize the weak texture and weak contrast structure of the cochlear alveolar region. As a result, the details of the alveolar region are visually submerged, statistically compressed, and structurally exhibit pseudo-boundaries or discontinuities.

[0023] Based on the above understanding, the core concept of this application is not to simply regard strong bony reflections as a background that needs to be darkened, but rather as an interference source with clear geometric boundaries and directional propagation characteristics. In the embodiments of this application, firstly, strong bony reflection areas are identified in the original imaging data and a bone reflection source map is formed, so that the strong reflection interface can be located in space; further, the local geometric features of the interface are extracted from the bone reflection source map, so that the directionality of the reflection influence has a calculable basis; and combined with the acquired parameter information, a contamination propagation characterization matching the interface geometry and distance relationship is established, so that the compensation of the alveolar region no longer relies on simple empirical parameter tuning or image-level enhancement, but can be controlled by eliminating the actual crowding effect of strong reflections on the neighborhood. Through this modeling approach based on the physical characteristics of the scene, this application aims to solve the dynamic range competition problem when strong reflection interfaces and weak signals in the deep alveolar region coexist in cochlear endoscopy imaging, thereby improving the recognizability of alveolar texture and weak contrast structures without destroying the authenticity of anatomical boundaries, and providing a more stable imaging basis for intraoperative observation, positioning assistance, or subsequent quantitative analysis.

[0024] refer to Figure 1 This is an exemplary application scenario diagram provided in the embodiments of this application.

[0025] like Figure 1 As shown, in the application scenario, the endoscopic imaging instrument, i.e., the endoscopic imaging cochlea, extends into the cochlear cavity along the entrance direction to capture images of the deep cochlear region. The cochlear cavity is schematically represented by a simplified spiral structure. Deep within the cavity, near the cavity wall, is the cochlear alveolar region. This alveolar region of the cochlea is within the field of view of the endoscopic imaging instrument, but it is relatively close to the cavity wall and is in an adjacent area that may be affected by strong reflective interfaces. Due to the curved or spiral structure of the cochlear cavity, the acquisition angle of the endoscopic imaging instrument is spatially limited, and the imaging optical path mainly enters the cavity from one side for illumination and reception.

[0026] refer to Figure 2 This is a schematic diagram of the endoscopic imaging cochlea provided in an embodiment of this application.

[0027] like Figure 2 As shown, the endoscopic imaging cochlea includes: A cochlear mold, comprising an insertion part and a probe end disposed at the distal end of the insertion part, the probe end being used to enter the cochlear cavity and perform near-field imaging of the cochlear alveolar region; An imaging component is disposed at the end of the probe. The imaging component includes an imaging sensor and an imaging optics corresponding to the optical axis of the imaging sensor, and is used to acquire raw imaging data of the alveolar region. An illumination component, disposed at the end of the probe and facing the framing direction, is used to provide illumination light to the cochlear alveolar region to support imaging by the imaging component; A processing component, electrically connected to the imaging component, is configured to generate a bone reflection source map based on the original imaging data, extract local geometric features of the bony strong reflection area, determine the relative distance information of the location to be processed relative to the bony strong reflection area, construct a bone reflection contamination propagation kernel by combining the acquired parameter information, and perform digital blanking compensation on the original imaging data based on the bone reflection source map and the bone reflection contamination propagation kernel to output the cochlear alveolar region imaging result; An output component, electrically connected to the processing component, is used to output the compensated cochlear alveolar bone region imaging results to a display terminal.

[0028] Next, with reference to the accompanying drawings, the high-definition imaging method for the implant bone groove region based on a cochlear implant provided in this application will be further elaborated. Figure 3 The methods shown include: S1: Acquire the raw imaging data of the alveolar region and the acquisition parameter information corresponding to the raw imaging data; In this embodiment, the raw imaging data is acquired by the imaging component of the endoscopic imaging device when it captures near-field images at the probe tip. The raw imaging data can be a single-frame image, a continuous image sequence, or short-segment sampling data composed of continuous frames, as long as it can at least reflect the grayscale distribution and structural boundaries within the cochlear cavity. This application does not impose further limitations. Acquisition parameter information is used to characterize the operating state of the imaging link in the current frame or current sampling segment, such as exposure, gain, emission energy, and scanning angle. Those skilled in the art will understand that acquisition parameters can be obtained from metadata recorded internally by the device, output from the drive end, or written synchronously with the image.

[0029] S2: Based on the original imaging data, identify the bony strong reflective areas in the alveolar region to obtain a bone reflective source map; The bone reflection source map at least characterizes the spatial distribution and boundary information of bony strong reflection regions. In this embodiment, the bony strong reflection region is not limited to a fully saturated bright spot, but refers to the high reflectivity response region and its boundary transition zone generated by the bony interface near the cochlear cavity wall. This region typically exhibits a significantly higher grayscale than its neighbors and continuous edges. To avoid misidentifying real bright textures within the alveolar region as reflection sources, the original imaging data is preferably subjected to noise reduction and grayscale normalization during the identification process. Candidate boundaries are formed by combining bright candidate region extraction and boundary detection, and then filtered based on connectivity and boundary continuity to determine the bony strong reflection region.

[0030] S3: Based on the bone reflex source map, extract the local geometric features of the bony strong reflex area; The local geometric features include at least the region normal direction and the region curvature. In this embodiment, the local geometric features are used to characterize the directionality and morphological changes of the strong reflection interface, avoiding the simplistic view of strong reflection as isotropic diffusion. Specifically, a sequence of boundary points of the bony strong reflection region can be extracted from the bone reflection source map, and several target boundary points can be selected as local representative points on the boundary point sequence. The selection of target boundary points can take into account both boundary coverage and the representativeness of reflection intensity. For example, based on equidistant sampling, the boundary reflection intensity value can be used for screening, so that key locations with high curvature or high reflection are not missed during subsequent modeling, while controlling the amount of computation. For each target boundary point, a local curve is fitted based on its neighborhood boundary point set to obtain the tangential direction and determine the region normal direction accordingly. At the same time, the reciprocal of the radius of curvature of the fitted curve characterizes the local curvature.

[0031] S4: Based on the local geometric features, determine the relative distance information of the location to be processed in the original imaging data with respect to the bony strong reflective area, and construct the bone reflection contamination propagation kernel by combining the acquired parameter information; The location to be processed refers to the pixel location in the original imaging data that needs to undergo bone reflection contamination assessment. In this embodiment, the location to be processed is not an arbitrary pixel within the bone groove area, but rather a pixel location in the neighborhood outside the strong reflective interface of the bone that is at risk of contrast compression or texture submersion due to strong reflection, in order to avoid blindly processing the entire image and causing the true structure to be mistakenly clipped. The relative distance information preferably uses the target boundary point as the distance reference point. A discrete sampling chain is generated in the normal direction outside the target boundary point, and the target pixel location is determined in the sampling chain according to rules such as grayscale attenuation inflection point, gradient directional peak, or local contrast valley. The sampling step size and sequence number represent the projection interval from the target pixel location to the target boundary point, thereby obtaining the relative distance information of the location to be processed relative to the strong reflective interface. The bone reflection contamination propagation kernel is used to characterize the directional and distance attenuation effects of the strong reflection source on the outer neighborhood. Its construction process simultaneously introduces local geometric features and acquisition parameter information, making the main diffusion direction of the kernel consistent with the normal / tangential coordinate system. The diffusion scale and attenuation intensity can be adjusted in conjunction with parameters such as exposure, gain, emission energy, and scanning angle.

[0032] S5: Based on the bone reflection source map and the bone reflection contamination propagation kernel, perform digital blanking compensation on the original imaging data to obtain the compensated cochlear alveolar region imaging result; In this embodiment, digital hidden surface compensation does not aim at simple global de-highlighting or single subtraction. Instead, under the constraint of the strong reflection interface located by the bone reflection source map, it performs limited compensation on the position to be processed, so as to reduce the encroachment of strong reflection on the neighboring grayscale space and restore the recognizability of weak contrast details in the bone groove area.

[0033] Before detailing the specific technical aspects of the steps, this application's embodiments need to reiterate: Near-field imaging in cochlear endoscopy is not simply a matter of insufficient brightness or random noise superposition. A more common challenge arises from the continuous encroachment of highly reflective interfaces on the signal carrying capacity of neighboring areas. Under near-field illumination, bony interfaces generate high-energy echoes or bright responses. This response not only exhibits local saturation at the interface but also, under the combined effects of the imaging link and tissue scattering conditions, creates band-like or fan-shaped grayscale increases and contrast compression in the outer neighborhood, crowding out the grayscale space originally used to characterize weak textures and low-contrast structures. This effect is distinctly directional and gradual, and extremely sensitive to acquisition parameters. Consequently, it exhibits an unstable bright area expansion-detail loss phenomenon under different exposures, gains, or probe postures. Simply using threshold suppression or global enhancement often only visually alleviates the bright areas but fails to stably restore the details of the alveolar region at the structural level.

[0034] Based on the above understanding, this embodiment does not treat strong reflections as abnormal bright spots that need to be simply erased, but rather as localizable interference sources with geometric boundaries. The impact of these interference sources on the neighborhood is described as a calculable propagation behavior. Unlike common deglare methods that directly perform equalization or filtering in the image domain, this embodiment prioritizes a set of interpretable conditions: where the influence originates, in what direction it spreads, how it attenuates with distance, and to what extent it is amplified under the current acquisition conditions. To this end, strong reflection interfaces are abstracted as a set of representative boundary locations. The local normal direction and curvature at the boundaries are used to characterize the main direction and morphological changes of the diffusion, the boundary reflection intensity provides a benchmark for the scale of the influence, and acquisition conditions such as exposure, gain, emission energy, and scanning angle serve as modulation factors, enabling the propagation behavior to adapt to different acquisition conditions.

[0035] It should be noted that this embodiment does not adopt a coarse-grained strategy of compensating the entire image point by point when determining the processing target. Instead, the compensation action is limited to the neighborhood location with a clear geometric relationship to the interference source. Specifically, the determination of the processing location does not rely on manually delineating the alveolar region, nor does it require prior knowledge of the fixed divisions of the cochlea. Instead, it uses the strong reflection boundary as a reference to form a discrete candidate sampling chain in the normal direction on its outer side, and searches for key locations in the sampling chain that can characterize the transition from the strong reflection-affected area to the normal area, such as the inflection point of grayscale decay, the peak of significant gradient directivity, or the valley of contrast compression.

[0036] Next, we will further elaborate on the technical content of the method of this application regarding the bony strong reflex area.

[0037] In one example, the identification method for the bony hyperreflex area includes: The original imaging data is preprocessed to obtain preprocessed imaging data. The preprocessing includes at least one of noise reduction processing and grayscale normalization processing. Specifically, the purpose of preprocessing is to stably highlight highly reflective interfaces from complex backgrounds, ensuring that subsequent candidate extraction is unaffected by speckle, dark quantization noise, or uneven local illumination. Noise reduction employs edge-preserving techniques such as bilateral filtering, guided filtering, or non-local mean filtering to preserve the gradient transition of cavity wall boundaries as completely as possible, while suppressing minor bright spots and random texture fluctuations caused by near-field illumination. Gray-level normalization eliminates overall brightness drift caused by exposure and gain differences between frames. In practice, image grayscale can be mapped to a unified target dynamic range, for example, linearly or piecewise linearly mapping the original grayscale values ​​to an 8-bit grayscale range of 0–255, or mapping to a 10-bit or 12-bit grayscale range of 0–1023 or 0–4095 under high dynamic range sensor output conditions. The specific target dynamic range selected depends on the bit depth of the imaging sensor and the grayscale accuracy requirements of subsequent processing modules; it is sufficient to ensure that highly reflective areas remain significantly higher than the grayscale of neighboring tissues after mapping, while retaining resolvable grayscale levels in weakly textured areas. Those skilled in the art will understand that grayscale normalization can employ linear stretching, piecewise linear mapping, or stretching strategies based on percentile truncation, as long as the strong reflective area remains significantly higher than the surrounding tissue after normalization. This application does not impose further limitations.

[0038] High-brightness candidate regions are extracted based on the preprocessed imaging data to obtain strong bony reflection candidate regions. Boundary detection is then performed on the strong bony reflection candidate regions to extract the candidate region boundary information. Specifically, in practice, the grayscale distribution within the entire image or a local sliding window can be statistically analyzed first. A high percentile grayscale value is selected as the initial threshold, and fine-tuned using local mean and local variance to adapt the threshold to variations in cavity depth. The resulting binary highlight mask after thresholding typically contains scattered noise and broken segments; therefore, morphological opening / closing operations or connected component filling operations can be introduced to enhance region continuity. Boundary detection can be performed on the candidate mask, for example, by extracting the outer contours of connected components through contour tracking, or by performing gradient edge extraction on the original grayscale image in conjunction with the candidate mask to obtain candidate boundary information that more closely resembles the real interface.

[0039] For example, to facilitate understanding of the above method of using high percentile grayscale values ​​as the initial threshold and fine-tuning them, the grayscale can be normalized to 0-255. First, the grayscale histogram of the entire image can be statistically analyzed, and the grayscale value corresponding to the 98th percentile can be used as the initial threshold. For instance, if the statistical analysis yields a grayscale value of 215 at the 98th percentile, then 215 can be used as the global initial threshold to obtain the first version of the binary high-brightness mask. Considering the variations in depth and uneven illumination within the cochlear cavity, a 31×31 or 41×41 local sliding window can be further used for threshold fine-tuning: if the local mean is 120 and the standard deviation corresponding to the local variance is approximately 18 within a certain sliding window, then the local threshold of that sliding window can be increased to around 225; if the local mean is 85 and the standard deviation is approximately 12 within another darker sliding window, then the local threshold of that sliding window can be decreased to around 205. Therefore, the threshold exhibits an adaptive variation range of approximately 205 to 225 at different spatial locations, which can both cover the bright areas of the bony, highly reflective interface and reduce the probability of dark noise being misjudged as strong reflection. The above values ​​are only illustrative values. Those skilled in the art can adjust the percentile selection and fine-tuning range according to the sensor depth, exposure gain range, and target image contrast level. This application does not impose further limitations.

[0040] Based on the connectivity features of the candidate regions for strong bony reflexes and the boundary information of the candidate regions, the candidate regions for strong bony reflexes are screened to determine the strong bony reflex regions. Understandably, connectivity features are used to constrain the topological rationality of candidate regions, while boundary information constrains their geometric rationality. Specifically, geometric quantities such as area, perimeter, thinness, and roundness can be calculated for each candidate connected region, and the boundary curvature variation can be combined to determine whether it has interface-type continuous features. Simultaneously, the proportion of bright pixels and grayscale uniformity within the connected region can be statistically analyzed to eliminate pseudo-connected regions formed by only a few isolated bright spots. To avoid misclassifying the real structure within the bone groove region as a strong reflection source, boundary gradient consistency judgment can be introduced during screening. This requires that the candidate region boundary has a stable gradient direction and a high gradient magnitude on the grayscale image, thus better conforming to the imaging rules of reflective interfaces.

[0041] It should be noted that the number of bony strong reflective regions in this application can be one or more. In the same frame of original imaging data, the bony interface may appear as a single continuous bright band, or it may appear as multiple separate strong reflective connected regions due to cavity curvature, occlusion, or differences in local bone wall morphology. The method of this application is based on the independent characterization and unified constraint processing of each strong reflective connected region in the bone reflective source image to carry out subsequent calculations. That is, for each strong reflective connected region, its boundary point sequence and target boundary point set are extracted, and the corresponding regional normal direction and regional curvature are determined respectively. The relative distance information and kernel parameter set are constructed using their respective target boundary points as distance references. On this basis, each strong reflective connected region can generate a corresponding bone reflective contamination propagation kernel, which is matched and called according to the distance reference relationship when evaluating and compensating the position to be treated, or the contributions of multiple connected regions are superimposed in a restricted manner to obtain a comprehensive effect.

[0042] In one example, the bone reflectance source map is a binary mask or labeled map obtained by stitching together at least one bony strong reflectance region, where different connected components correspond to different bony strong reflectance regions. The bone reflectance source map can use the same spatial resolution and pixel coordinate system as the original imaging data to identify the location range and boundary orientation of the strong reflectance regions at the pixel level. Those skilled in the art will understand that the stitching can be understood as uniformly mapping and superimposing the selected strong reflectance connected components to form a complete spatial distribution representation within the same image matrix, without changing the size or sampling structure of the original image.

[0043] Specifically, the bone reflection source map is used for subsequent local geometric feature extraction and distance reference establishment. On the one hand, the boundary point sequence of each connected region can be directly extracted from the bone reflection source map, providing a clear boundary trajectory for the calculation of the region's normal direction and curvature. On the other hand, the bone reflection source map is also used to limit the search range of the location to be processed, so that the determination of subsequent relative distance information only revolves around the outer neighborhood of the bone reflection source map, avoiding irrelevant calculations at locations far from strong reflection areas.

[0044] Next, we will further elaborate on the technical content of the method of this application regarding local geometric features.

[0045] In one example, the extraction of the local geometric features includes: S3.1: Extract the boundary point set of the strong bony reflection region in the bone reflection source map to obtain the boundary point sequence of the strong bony reflection region; Specifically, the boundary point sequence must not only delineate the region but also maintain boundary continuity and traceability at the pixel level, avoiding directional jitter caused by breaks, burrs, or jagged edges. Therefore, boundary extraction is not simply about finding the contour of the mask, but rather about first performing minimal morphological consistency adjustments on the mask. For example, closing operations are performed using structuring elements of 1 to 2 pixels to fill small holes, and then opening operations are used to suppress isolated noise points, making the boundary closer to the continuous shape of the real interface in the image, thus providing stable input for subsequent local fitting.

[0046] In this embodiment, boundary point extraction can be achieved using connected component contour tracking: for each bony, highly reflective connected component, its outer contour starting point is first located, such as the topmost or leftmost boundary pixel. Then, it is tracked point by point along the boundary according to a preset neighborhood search rule, outputting a sequentially arranged sequence of boundary points. To ensure that the boundary point sequence is consistent in direction, the contour output direction can be uniformly defined, such as clockwise or counterclockwise, and the pixel coordinates of each boundary point and its index in the sequence are recorded. If there are multiple connected components, a separate boundary point sequence is output for each connected component, and a connected component identifier is appended to the beginning of the sequence, so that subsequent calculations can be performed independently by connected component without cross-component aliasing.

[0047] S3.2: Select a target boundary point from the boundary point sequence, and determine a set of boundary neighborhood points with the target boundary point as the center of a preset neighborhood range; Specifically, boundary point sequences typically contain a large number of points. Calculating the tangent and curvature for each point would impose a computational burden, and local pixel jitter at the boundary could lead to instability in orientation estimation. Therefore, the selection of target boundary points needs to balance covering the main orientation of the boundary and highlighting parts representative of the propagation pattern. Furthermore, in this method, it is also necessary to provide reliable reference positions and intensity parameter sources for the subsequent kernel parameter set. Based on this consideration, target boundary points should not be randomly extracted but should be determined through regular sampling and screening to ensure that they reflect the overall boundary morphology and have higher density in locations with stronger reflectivity and more significant curvature changes.

[0048] In one example, the target boundary points are selected as follows: the boundary point sequence is sampled at equal intervals according to a preset sampling interval to obtain an initial sampled boundary point set; for each initial sampled boundary point in the initial sampled boundary point set, the boundary reflection intensity value of the initial sampled boundary point is extracted from the original imaging data, the boundary reflection intensity value is determined at least by the pixel grayscale value at the initial sampled boundary point and the proportion of bright pixels in its neighborhood; the initial sampled boundary point set is subjected to intensity-weighted filtering based on the boundary reflection intensity value, so as to select the initial sampled boundary point whose boundary reflection intensity value meets the preset intensity condition as the target boundary point, and the boundary reflection intensity value is used as the kernel intensity parameter of the corresponding target boundary point.

[0049] Specifically, the preset sampling interval for equidistant sampling can be set by combining the length of the boundary point sequence and the image resolution, so that the target boundary points can cover the overall direction of the strongly reflective interface without causing a burden on subsequent fitting and kernel parameter calculation due to excessive point density. Specifically, let the length of the boundary point sequence be N. When N is in the common range of 800 to 1200, the sampling interval can be set to 10 to 15; when N is short, the sampling interval can be reduced to 6 to 8 to avoid missing local bending segments; when N is long, the sampling interval can be increased to 16 to 20 to control the number of points. For example, if the length of the boundary point sequence of a certain bony strongly reflective connected region is 1000, the sampling interval can be set to 12, thus obtaining an initial set of approximately 83 sampled boundary points. To ensure coverage of the beginning and end, the starting index can be set to 0 and points can be taken incrementally at intervals. Finally, if the remaining segment at the end exceeds half an interval, the end boundary point is added to the initial sampling set to avoid discontinuity in direction estimation caused by the shortcomings of the boundary end. This setting method does not rely on fixed cavity partitions and can adaptively adjust to changes in the length of strong reflection boundaries under different fields of view.

[0050] In this embodiment, the boundary reflection intensity value is used to characterize the intensity scale of the boundary location as an interference source. Its calculation should not rely solely on the grayscale of a single pixel to avoid distortion caused by noise or single-point saturation. Specifically, for each initial sampling boundary point, a neighborhood window is extracted from the original imaging data centered on the pixel coordinates of that point. The window size can be exemplarily set to 9×9 pixels. Within this window, the grayscale value of the center pixel is recorded. And calculate the proportion of bright pixels. The threshold for determining the highlight pixel can be determined by the overall grayscale distribution of the image. An example of how to determine the threshold is as follows: Calculate the grayscale histogram of the current frame and take the grayscale value corresponding to the 95th percentile as the threshold. For example, after normalization to 0-255, if the 95th percentile grayscale value is 200, then pixels with a grayscale value greater than or equal to 200 within the window are counted as highlighted pixels, and the number of highlighted pixels is recorded as follows: If the total number of pixels is 81, then for Divide by 81. Boundary reflection intensity value can be and By combining, for example, to form a normalized intensity quantity through weighted synthesis: first, Normalize 0 to 255 to 0 to 1 to obtain , The normalized center pixel grayscale value is then adjusted according to a preset weight. and For synthesis, exemplary weights can be 0.7 and 0.3, making the central grayscale contribution slightly higher than the proportion contribution; when it is necessary to suppress the misleading effect of single-point saturation on intensity, the weights can be... Replace with the upper quartile mean of the grayscale values ​​within the window or the mean after removing extreme values, thus... This better reflects the realism of the "continuous highlight" near the interface. Those skilled in the art will understand that the aforementioned window size, percentile threshold, and weights can be adjusted according to the imaging resolution and exposure gain range, as long as it ensures that different boundary points within the same frame... As long as they are comparable, that's sufficient.

[0051] Furthermore, intensity-weighted filtering is used to retain representative points from the initial set of sampled boundary points that are more likely to dominate the neighborhood crowding effect, while avoiding over-biasing due to a few abnormally high-brightness points. Specifically, all initial sampled boundary points can be filtered first. Statistical analysis is performed to obtain the mean and standard deviation, or the median M and interquartile range (IQR). In a more robust implementation, the median M can be used as the central metric, and the IQR can be used to determine the degree of dispersion. A preset strength condition can then be set as "Ib is not less than M plus k times IQR," where k can be 0.5 to 1.0. For example, if the median Ib M of a connected component is 0.62 and the IQR is 0.10, then a threshold of 0.67 to 0.72 can be used. When k is 0.8, the threshold is 0.70, and all initial sampling boundary points with Ib greater than or equal to 0.70 are determined as target boundary points. To avoid excessive clustering of target boundary points on the boundary, a spatial uniformity constraint can be added: if the index spacing between two adjacent candidate target boundary points on the boundary sequence is less than a preset minimum spacing (e.g., 6 boundary points), then only the point with the larger Ib is retained, and the other point is removed, thus ensuring a more uniform distribution of target boundary points along the boundary.

[0052] S3.3: Based on the set of boundary neighborhood points, perform curve fitting on the local boundary of the bony strong reflective region to obtain the tangential direction at the target boundary point, and determine the regional normal direction at the target boundary point according to the tangential direction; Specifically, the tangential and normal directions are used to establish local coordinates based on the boundary, enabling subsequent propagation patterns to distinguish between diffusion along the interface and diffusion along the outer direction. Since boundary points come from the pixel grid, directly using the difference between adjacent points can easily cause directional jumps, leading to instability in the subsequent main diffusion direction. Therefore, the introduction of curve fitting is not to pursue high-order geometric accuracy, but to obtain a smooth local boundary approximation at the boundary neighborhood scale, thereby stably defining the tangential and normal directions on this approximation and ensuring consistency in direction calculations across different frames and noise levels.

[0053] In this embodiment, curve fitting can be achieved using low-order polynomial fitting or local spline fitting. Taking low-order polynomial fitting as an example, the coordinates of the points in the boundary neighborhood set can be mapped to a local coordinate system with the target boundary point as the origin. Then, the projection of the neighborhood point set on the principal direction is selected as the independent variable for fitting to reduce numerical instability caused by the vertical slope. The fitting order can be selected as second or third order, to balance the curve bending expression and the suppression of overfitting. After fitting, the local tangent direction of the fitted curve at the target boundary point is taken as the tangential direction, and the tangential direction is normalized. The region normal direction can be obtained by rotating the tangential direction, and the outer normal direction needs to be determined by the inner and outer side judgment: the bone reflection source map mask can be used to determine whether the normal side falls inside the strong reflection area. If it falls inside, the opposite direction is taken as the outer normal direction, thereby ensuring that the outer normal direction always points to the side of the non-bone strong reflection area, which is convenient for establishing a sampling chain and distance reference along the outer normal direction.

[0054] S3.4: Determine the local curvature at the target boundary point based on the curve fitting results, wherein the local curvature is characterized by the reciprocal of the radius of curvature formed by the set of boundary neighborhood points; Specifically, local curvature reflects the degree of boundary bending. In this method, its role is not only geometric description but also characterizing the risk of morphological changes caused by strong reflections: at locations with stronger curvature, near-field illumination and viewpoint occlusion are more likely to form non-uniform neighborhood uplift bands, and the propagation pattern is more likely to exhibit fan-shaped or oblique diffusion. Therefore, the calculation of local curvature needs to be defined at the same scale as the tangential and normal directions and associated with the target boundary points so that it can be directly referenced when determining subsequent directional morphological parameters.

[0055] In this embodiment, the local curvature can be indirectly obtained by the degree of curvature of the fitted curve at the target boundary point. To avoid introducing complex mathematical expressions, a geometric equivalent method can be used: three points are selected from the fitted curve or the set of neighborhood points to form a local arc segment. The three points can be the target boundary point, its neighboring point q steps forward, and its neighboring point q steps backward. For example, q is taken as 5 to 10. The three points determine an approximate arc, and the radius of curvature of the arc is calculated. The curvature is represented by the reciprocal of the radius of curvature, so that the smaller the radius, the larger the curvature. To ensure stability, when the three points are approximately collinear, resulting in an excessively large radius or unstable value, the lower limit of curvature can be constrained to a value close to zero, and the point can be treated as an approximate straight line segment in subsequent operations. For example, when the calculated radius of curvature is greater than 200 pixels, the curvature can be set to a low curvature range below 0.005 to reflect that the boundary at that point is approximately a straight line rather than a bend.

[0056] S3.5: The normal direction of the region and the curvature of the region are summarized at multiple target boundary points of the bony strong reflective region to obtain the local geometric features of the bony strong reflective region; Specifically, bony regions with strong reflectivity are not single-point geometric objects, but rather a collection of interfaces. The propagation effects of these interfaces may exhibit directional biases and morphological differences at different boundary locations. If only a single normal or a single curvature is used to describe the entire connected domain, deviations are easily introduced at long or curved boundaries, resulting in subsequent propagation characterization failing to cover local differences.

[0057] In this embodiment, the aggregation method can be expressed using a structured data table or vector set: for each target boundary point, its pixel coordinates, boundary reflection intensity value, tangential direction, outer normal direction, and local curvature are recorded, and these records are sorted by boundary sequence index to form a list of geometric features of the target boundary points. For the case of multiple connected components, multiple lists are formed and the connected component identifiers are retained to avoid mixing direction information across connected components. To facilitate subsequent interpolation along the boundary, a consistency constraint can be imposed on the normal direction between adjacent target boundary points: if the change in the normal angle between two adjacent points exceeds a preset threshold, for example, 45 degrees, a supplementary boundary point can be inserted between the two points and the direction can be re-estimated through local fitting, or spherical linear interpolation can be performed on the normal direction to make the direction change smoother and reduce abrupt changes in the main diffusion direction of the kernel at the boundary.

[0058] Next, we will further elaborate on the technical content of the bone reflex contamination transmission nucleus in this application.

[0059] It should be noted that the bone reflection contamination propagation kernel is not used for simple smoothing or weakening of highlights, but rather to characterize the propagation law of the non-target intensity contribution introduced by the bony strong reflective interface to the outer neighborhood under the current acquisition state. This allows subsequent compensation to be carried out based on a set of interpretable conditions: where the interference comes from, in what direction it spreads, how it attenuates with distance, and under what acquisition conditions it is amplified. Because the influence of the strong reflective interface in cochlear endoscopy near-field imaging is directional and gradual, and coupled with acquisition parameters such as probe posture and exposure gain, if only a fixed template or a single-scale suppression strategy is used, situations often arise where some frames are insufficiently suppressed, resulting in the alveolar region still being covered by grayscale enhancement, while other frames are excessively suppressed, leading to the accidental clipping of true textures. By using the propagation kernel as a propagation representation that can adaptively change with geometry and parameters, the controllability of the neighborhood crowding effect can be maintained under different viewing angles and brightness conditions, thereby restoring the details of the alveolar region to a more stable and consistent state.

[0060] Before detailing the bone reflection contamination propagation kernel, it's necessary to clarify the role and limitations of the processing location to understand the necessity of relative distance information in subsequent modeling. It's important to understand that the processing location is neither inside the bony strong-reflection area nor equivalent to the boundary point of the strong-reflection area itself. The processing location corresponds to a pixel position in the outer neighborhood of the bony strong-reflection area that is more likely to exhibit increased grayscale, compressed contrast, or submerged texture in the imaging results. The interior of the strong-reflection area often contains saturated or near-saturated interface responses, and its pixel grayscale reflects illumination and interface reflection conditions rather than the structural information of the bone groove area. Numerical suppression of this area cannot directly recover the details of the bone groove area. Conversely, what truly affects the discernibility of the bone groove area is usually the spatial encroachment effect of non-target intensity contributions introduced by the strong-reflection interface into the outer neighborhood. This encroachment effect manifests as increased grayscale in the neighborhood and compressed grayscale range, resulting in insufficient contrast carrying capacity for weakly reflective textures near the cavity wall.

[0061] It should be noted that the relative distance information is used to construct the bone reflection contamination propagation kernel not by calculating the propagation kernel and then inferring the distance, but because in the imaging mechanism framework of this embodiment, the distance itself is a priori geometric quantity, and its determination is independent of the propagation kernel. Moreover, it can provide a measurable independent variable for the part that affects how the spatial decay occurs before the propagation kernel is established. The location and boundary orientation of the bony strong reflective interface in the image can be directly given by the bone reflection source map. Based on the outer normal direction at the boundary, a spatial scale is established from the interface outward. The obtained relative distance reflects the degree to which the position to be processed leaves the interface along the main influence path. In other words, the relative distance provides the coordinates and scale reference of the propagation kernel, enabling the propagation kernel to maintain a decay law consistent with the spatial position under different frames and different exposure gain conditions, rather than treating the distance as an unknown quantity that needs to be interpreted by the propagation kernel.

[0062] In one example, the relative distance information is determined in the following ways: For each target boundary point in the bony strong reflective region, the outer normal direction is determined based on the regional normal direction at the target boundary point, and a discrete sampling chain is generated along the outer normal direction starting from the target boundary point. The discrete sampling chain consists of multiple candidate pixel positions obtained by expanding outward point by point according to a preset sampling step size. Specifically, the generation of the discrete sampling chain is based on the geometric constraint that the outer normal direction is the main influencing path, ensuring that the definition of relative distance is consistent with the gradual encroachment of the neighborhood by the strong reflective interface. The determination of the outer normal direction depends not only on the normal vector obtained by tangential rotation but also on the distinction between inner and outer sides based on the bone reflection source map: starting from the target boundary point, several pixels are probed along the positive and negative normal directions respectively, for example, a step size of 1 to 3 pixels. If the probe point still falls within the bone strong reflective mask, this direction is determined as the inner direction, and its reverse direction is taken as the outer normal direction, thus ensuring that the sampling chain always expands outward into the non-strong reflective area. The step size and length of the discrete sampling chain can be set according to the image resolution and cavity scale. For example, the preset sampling step size can be 1 pixel unit to obtain continuous distance scales, and the maximum outward expansion step size can be 30 to 80 steps to cover the transition area from the adjacent interface to the relatively stable background. To avoid interference from saturated pixels at the boundary to subsequent index calculations, an initial offset can be set. This allows the sampling chain to start recording candidate pixels from a position a certain distance from the boundary. For boundary segments with significant curvature, the sampling chain may cross into adjacent connected regions or enter occluded areas during its outward expansion. At each step, it can be determined whether the candidate pixel is still in the effective field of view or has entered other strong reflection masks. Once a boundary crossing or entry into other strong reflection areas occurs, the chain is truncated to ensure the single-source consistency of the distance reference.

[0063] In the discrete sampling chain, at least one target pixel position corresponding to the position to be processed is determined based on a preset point selection rule. The preset point selection rule includes at least the inflection point determination of the grayscale attenuation index, the peak value determination of the gradient magnitude under the condition that the gradient direction points to the target boundary point, and the valley value determination of the local contrast compression index. Specifically, the introduction of preset point selection rules is not to find the brightest or darkest point, but to locate key positions on the sampling chain that represent the transition from strong to weak reflection effects, ensuring a stable and consistent correspondence between the position to be processed and the distance reference. The grayscale attenuation index can be calculated point by point on the sampling chain. For example, by recording the sequence of changes in candidate pixel grayscale with the number of expansion steps, and looking for inflection points in this sequence where the change transitions from slow to rapid or from rapid to stable. In practice, the change amplitude of adjacent differences can be used to determine the inflection point position. For example, when the grayscale difference of three consecutive steps jumps from less than 5 to greater than 12, the position near this jump point is taken as a candidate inflection point. The gradient magnitude peak determination is used to capture the leading edge of strong reflection trails or halos in space. It calculates the gradient magnitude and estimates the gradient direction at each candidate pixel in the sampling chain, then determines whether the gradient direction points to the target boundary point. Using the line connecting the target boundary point to the candidate pixel as a reference, if the angle between the gradient direction of the candidate pixel and this reference direction is less than a preset angle threshold (e.g., 20 to 30 degrees), the directional condition is considered met. The point with the largest gradient magnitude in the set of points meeting the condition is selected as the peak position. The local contrast compression index is used to locate areas where texture carrying capacity is squeezed. It calculates the grayscale variance or gradient energy as the contrast value within the candidate pixel neighborhood (e.g., a 7×7 window), and searches for the valley point of the contrast value along the sampling chain. To avoid misjudging accidental low variance caused by noise as a valley point, the contrast value at the valley point can be required to be lower than the median contrast value in the chain by a certain proportion (e.g., 0.6 to 0.8), with a significant upward trend several steps before and after it.

[0064] Establish a mapping relationship between the target pixel position and the corresponding target boundary point, and use the corresponding target boundary point as the distance reference point of the target pixel position; Based on the projection interval between the target pixel position and the corresponding target boundary point in the outer normal direction, the relative distance information of the target pixel position relative to the bony strong reflective region is determined, wherein the projection interval is characterized by the preset sampling step size and the index of the target pixel position in the discrete sampling chain. Specifically, determining the mapping relationship and projection interval is used to convert the selected target pixel position into a distance dimension that can be used for kernel attenuation modeling. The mapping relationship can be recorded using indexing: assign a boundary point number to each target boundary point and record the corresponding sampling chain sequence; when the target pixel position is determined on the sampling chain, record the target pixel's index in the sampling chain, and write the boundary point number - index - pixel coordinates as a mapping entry into the pending position table. The projection interval can be calculated directly using the sampling step size and index, i.e., the projection interval equals the index multiplied by the sampling step size; for example, if the sampling step size is 1 pixel, and the target pixel index is 18, then the projection interval is 18 pixels. If the sampling chain uses a starting offset... Then the relative distance can be further increased by this offset, for example... If the distance is 2 pixels, the relative distance is 20 pixels to maintain a complete distance scale starting from the outer edge of the boundary. For multi-connected components, each connected component maintains its own mapping entry and distance reference to avoid the same target pixel being repeatedly associated with multiple boundary points, leading to distance uncertainty. When a target pixel falls within the reachable range of two boundary point sampling chains simultaneously, the boundary point with the smaller projection interval can be preferentially selected as the distance reference point, or the boundary point with the higher boundary reflection intensity value can be selected as the reference point to ensure that the distance reference is consistent with the main interference source. Those skilled in the art will understand that the aforementioned angle threshold, difference threshold, window size, and scaling threshold can all be adjusted according to the imaging resolution and noise level, as long as the distance calculation within the same frame has a consistent scale and comparability.

[0065] In one example, a bone reflex contamination transmission kernel is constructed by combining the collected parameter information, including: S4.1: Extract a set of nuclear modulation parameters to characterize the degree of strong reflection response expansion based on the acquired parameter information. The set of nuclear modulation parameters includes gain parameters, exposure parameters, emission energy parameters, and scanning angle parameters. Specifically, the neighborhood grayscale rise width and trailing pattern of a bony, highly reflective interface are inconsistent under different exposure, gain, emission energy, and scanning angle conditions. If the kernel parameters are fixed for a long period, the same interface may be constrained to varying degrees in different frames, causing the compensation scale to drift over time. In this embodiment, the acquired parameter information is preferably acquired synchronously with the original imaging data frame by frame or by sampling segment, and converted into a comparable scalar form for subsequent use in the coordinated adjustment of scale, attenuation, and orientation offset.

[0066] In this embodiment, the gain parameter can be directly read from the analog or digital gain recorded by the imaging device; the exposure parameter can be read from the exposure time or equivalent exposure level; the emission energy parameter can be read from the illumination brightness duty cycle, emission power level, or drive current setting; and the scanning angle parameter can be read from the pitch / yaw angle output by the probe attitude sensor, or from the angle encoding value of the scanning mechanism. When a parameter is given in discrete levels, it can be mapped to a continuous scale, for example, the gain levels 1 to 16 can be linearly mapped to a normalized range of 0 to 1, so that different parameters have a unified dimension during modulation. To avoid kernel modulation jitter caused by single-frame parameter fluctuations, the acquired parameters can be smoothed with a short window, for example, the median of the parameters of the most recent 5 frames can be used as the kernel modulation input of the current frame, so that the modulation remains continuous as the parameters change.

[0067] S4.2: Based on the local geometric features, determine the directional morphological parameters and main diffusion direction of the bone reflection pollution propagation nucleus, wherein the directional morphological parameters include diffusion scale parameters along the normal direction of the region and diffusion scale parameters along the tangential direction of the region. Specifically, the introduction of directional morphological parameters and the main diffusion direction stems from the spatial characteristic that strong reflection influences more significantly encroach along the outer direction and diffuse less along the boundary direction. Furthermore, this characteristic is more prone to skewness and width variations at curved boundaries. The local geometric features, given in the previous steps, include the outer normal direction, tangential direction, and local curvature, enabling the propagation kernel to express anisotropy within the local coordinate system, rather than approximating the neighborhood influence with uniform diffusion of a single radius.

[0068] In this embodiment, the diffusion scale parameters along the normal direction and along the tangential direction of the region can be determined by a base scale and curvature correction. The base scale can be empirically set based on image resolution and typical contamination band width. For example, under the condition of 0-255 gray levels and a resolution of approximately tens of micrometers per pixel, the normal base scale can be exemplarily set to 12 pixel units and the tangential base scale to 6 pixel units to reflect the more prominent normal encroachment. The curvature correction can adjust the scale according to the local curvature. When the curvature is large, the normal scale is appropriately increased to cover the outer width of the skewed diffusion band, while the tangential scale is slightly adjusted to conform to the boundary direction change. For example, the median can be taken from the set of curvature values ​​as the reference curvature. If the curvature of a target boundary point is higher than 1.5 times the reference curvature, the normal scale is increased by 3 to 5 pixel units on the base scale; if the curvature is lower than 0.7 times the reference curvature, the normal scale remains at the base scale or is decreased by 1 to 2 pixel units. The values ​​above are illustrative for ease of understanding. Those skilled in the art can adjust them according to the device's field of view and resolution, as long as the normal scale shows a consistent increasing trend with the increase of curvature. The main diffusion direction is recorded with the outer normal direction as the reference direction, so that subsequent directional offsets can be rotated and corrected around this reference.

[0069] S4.3: Based on the relative distance information, determine the distance attenuation parameter of the bone reflection contamination propagation core; Specifically, the relative distance information has established a mapping relationship between the target boundary point, the outer normal sampling chain, and the position to be processed in the previous steps, making the distance a priori variable independent of the kernel; the distance attenuation parameter characterizes the extent to which the contribution of non-target intensity introduced by strong reflection should be attenuated at this distance.

[0070] In this embodiment, the distance attenuation parameter can be characterized by two quantities: the effective influence radius and the attenuation slope level. The effective influence radius can be automatically estimated based on the stable segments of grayscale and contrast on the discrete sampling chain: scanning along the sampling chain from near to far, when the grayscale change amplitude is lower than the threshold and the local contrast no longer decreases for several consecutive steps (e.g., 8 steps), the distance corresponding to that position is regarded as the boundary where the influence tends to stabilize, and the effective influence radius is determined accordingly. The threshold can be determined using an intra-frame adaptive method: the median of the grayscale difference in the sampling chain is used as the background fluctuation scale. For example, if the median of the grayscale difference is 3, then the grayscale change amplitude threshold can be taken as twice the median, i.e., 6; the local contrast stabilization threshold can be taken as 0.9 times the median of the contrast in the sampling chain as the criterion for no further decrease. Assuming that a sampling chain starts to meet the condition of "the grayscale difference is less than 6 for 8 consecutive steps and the contrast no longer decreases" at a distance of 24 pixels, then the effective influence radius can be taken as 24 pixels. The attenuation slope level can be determined by combining the emission energy and exposure gain level: when the emission energy is high and the exposure gain is high, the attenuation is slower; when the emission energy is low or the exposure is low, the attenuation is steeper, thus making the nuclei more concentrated in the near-interface region.

[0071] S4.4: Adjust the directional morphological parameters and the distance attenuation parameters according to the nuclear modulation parameter set to generate a bone reflection contamination propagation nucleus; Specifically, the adjustment process maps changes in the acquisition state to linked changes in the amplitude, scale, direction, and attenuation of the kernel, ensuring that the propagation kernel corresponding to the same bony interface remains consistent with the actual degradation morphology in different frames and poses. Without this linkage, the parameters may remain unchanged, leading to insufficient suppression in some frames and excessive suppression in others, resulting in unstable visual and statistical performance of the compensation results.

[0072] In one example, adjusting the directional shape parameter and the distance attenuation parameter includes: A kernel intensity adjustment coefficient is determined based on the gain parameter and the exposure parameter, and the diffusion scale parameter along the normal direction of the region and the diffusion scale parameter along the tangential direction of the region are synchronously amplified or reduced according to the kernel intensity adjustment coefficient. A directional offset angle is determined based on the scanning angle parameters, and the main diffusion direction of the bone reflection contamination propagation nucleus is rotated and corrected according to the directional offset angle. An attenuation adjustment coefficient is determined based on the emitted energy parameters, and the distance attenuation parameters are adjusted according to the attenuation adjustment coefficient. Bone reflection contamination propagation nuclei are generated based on the adjusted diffusion scale parameters along the regional normal direction, the adjusted diffusion scale parameters along the regional tangential direction, the adjusted main diffusion direction, and the adjusted distance attenuation parameters.

[0073] It is understood that the aforementioned adjustment process can be implemented in engineering using existing parameter normalization and rule mapping techniques. For example, the determination of the nuclear intensity adjustment coefficient can be achieved using linear normalization based on the equipment calibration range, piecewise linear mapping, or lookup table mapping to convert the gain parameter and exposure parameter into a modulation factor under a unified dimension. The synchronous amplification or reduction of the diffusion scale parameter can be achieved using proportional scaling and upper / lower limit clamping strategies to ensure that the scale remains continuous with the modulation factor and does not exceed a preset reasonable range. The rotation correction of the main diffusion direction can be achieved through two-dimensional vector rotation, offset superposition based on the direction angle, or direction transformation based on the attitude matrix to apply the offset corresponding to the scanning angle parameter to the principal axis direction of the local coordinate system. The adjustment of the distance attenuation parameter can be achieved through parameterized adjustment of the exponential attenuation curve, piecewise attenuation strategy, or attenuation coefficient lookup table strategy to ensure that the attenuation trend remains consistent with the change of emission energy. Those skilled in the art will understand that the above implementation methods are mature numerical calculation and parameter modulation techniques that can complete the linkage adjustment of nuclear parameters without changing the overall modeling logic of this embodiment, and will not be elaborated upon here.

[0074] In one optional implementation, the process of determining the attenuation adjustment coefficient based on the emission energy parameters includes: firstly, acquiring the emission energy parameters under the current scanning conditions, and then mapping the emission energy parameters to the corresponding attenuation adjustment coefficients by combining the energy-attenuation response correspondence established in advance during the equipment calibration phase. The correspondence can be obtained by statistically fitting the pollution diffusion range, pollution intensity attenuation rate, and effective influence distance around the bony strong reflective area at different emission energy levels. Generally, the higher the emission energy, the larger the propagation range of the reflected signal in space, and the slower the pollution contribution attenuates with distance; therefore, a larger attenuation adjustment coefficient is determined accordingly. Conversely, the lower the emission energy, the more concentrated the pollution contribution is in the vicinity of the reflection source, and the faster it attenuates with increasing distance; therefore, a smaller attenuation adjustment coefficient is determined accordingly.

[0075] In another example, the bone reflection contamination propagation kernel is characterized by a set of kernel parameters associated with each target boundary point of the strong bone reflection region. The set of kernel parameters includes a kernel intensity parameter, a diffusion scale parameter along the normal direction, a diffusion scale parameter along the tangential direction, a main diffusion direction parameter, and a distance attenuation parameter. Based on the set of kernel parameters, a kernel weight function is determined in a local coordinate system with the corresponding target boundary point as the origin and formed by its tangential and normal directions.

[0076] Next, we will further elaborate on the technical content of the digital blanking compensation method in this application.

[0077] In one example, the compensation method for digital blanking compensation includes: Based on the bone reflectance source map and the bone reflectance contamination propagation kernel, a contamination contribution field of the outer neighborhood is generated for each target boundary point of the strong bony reflectance region in the corresponding local coordinate system to obtain the upper bound information of the contamination contribution of each outer neighborhood position. For the location to be processed, the limited compensation amount for the location to be processed is determined based on the relative distance information relative to the corresponding target boundary point and the pollution contribution upper bound information. A normal sampling chain consistency constraint is constructed based on the outer normal direction associated with the target boundary point, and the constrained compensation amount is corrected for normal consistency based on the normal sampling chain consistency constraint. The corresponding unprocessed position is compensated based on the limited compensation amount after normal consistency correction.

[0078] Understandably, the pollution contribution field is used to translate the spatially propagable effects of bony, highly reflective interfaces from a conceptual level into a computable data structure, so that the values ​​of subsequent compensation quantities have clear upper bound constraints rather than relying on empirical strength.

[0079] Specifically, for each target boundary point, a local coordinate system is first established with the target boundary point as the origin and the tangential direction and the outer normal direction as the boundary. An outer neighborhood calculation window is selected within this coordinate system as the support domain of the contribution field. The window range can be set in conjunction with the aforementioned effective influence radius, for example, covering 0 to 30 pixels in the normal direction and -15 to +15 pixels in the tangential direction. For each discrete position within the window, the kernel weight function is called based on the kernel parameter set corresponding to the target boundary point to obtain the pollution weight value at that position. This is then combined with the kernel intensity parameter of the target boundary point to generate a pollution contribution value, thus forming a "position-contribution value" contribution field matrix. To avoid unreasonable amplification caused by the superposition of contributions from multiple target boundary points, a restricted aggregation rule is preferably adopted in the summarization stage of the contribution field: when the same pixel position falls into multiple contribution field support domains simultaneously, the maximum contribution value is taken as the upper bound, or the upper bound value after distance weighting is taken and a saturation upper limit is set. For example, the upper limit can be set to a certain proportion of the dynamic range of grayscale in that frame, ensuring that the contribution upper bound does not exceed the bearable range of the original grayscale. This treatment allows the pollution contribution field to cover multiple sources of impact while maintaining the stability of its upper bound, facilitating the determination of subsequent limited compensation amounts.

[0080] In this embodiment, the limited compensation amount is not a simple subtraction of the original grayscale, but rather a compensation range given within an interpretable range by combining relative distance information and the upper bound of contribution. This allows the compensation to decrease naturally with distance from the interface and avoids over-correction in distant areas. Specifically, for each location to be processed, its corresponding distance reference point and outer normal sampling chain number are first determined through a mapping relationship to obtain the relative distance; then, the upper bound value of the contribution corresponding to the location to be processed is read from the contamination contribution field and used as the upper limit of the compensation range. The initial value of the compensation amplitude can be attenuated and allocated according to distance. For example, when the relative distance is within the effective influence radius, a certain proportion can be allocated from the upper bound according to the trend of stronger near-far and weaker far-far, with a higher allocation proportion for closer distances and a lower allocation proportion for farther distances. For example, the effective influence radius can be divided into three segments: near field, mid field, and far field. In the near field segment, 0.8 to 1.0 of the upper bound can be used as the initial compensation amount, in the mid field segment, 0.4 to 0.8, and in the far field segment, 0.1 to 0.4. Non-negativity and dynamic range constraints are applied to the compensation amount to avoid grayscale exceeding the limit or reverse enhancement after compensation. If a location to be processed is associated with multiple target boundary points, the boundary point with the smallest projection distance can be used to contribute the upper bound first, or a restricted merging value can be taken for multiple upper bounds to ensure that the compensation reference is consistent with the main interference source. Those skilled in the art will understand that the above segmentation ratio and radius division can be adjusted according to the resolution and noise level, as long as the compensation amount remains monotonically unchanged with increasing distance.

[0081] Furthermore, the normal sampling chain consistency correction is used to eliminate the local discreteness caused by point-by-point compensation, ensuring a continuous transition of the compensated grayscale along the outer normal direction and avoiding the formation of strip discontinuities or local reverse fluctuations. Specifically, for each outer normal sampling chain of the same target boundary point, the initial compensation amounts of all positions to be processed on the chain are arranged in distance order to construct a compensation amount sequence. Consistency constraints are applied to this sequence to ensure that the compensation amount decreases smoothly from the near boundary to the far boundary and maintains an interpretable transition shape near the inflection point. In practice, a combination of restricted smoothing and monotonic correction can be used: first, short-window smoothing is performed on the sequence to remove single-point anomalies, and then monotonic correction is performed to ensure that the compensation amount far from the interface is not higher than the compensation amount closer to the interface; if a small number of points have abnormally large compensation amounts due to local texture, they can be truncated to the interpolated values ​​of adjacent points, thereby maintaining the continuity of the chain. After completing the normal consistency correction, the corrected compensation amount is applied point by point to the grayscale correction of the corresponding position to be processed, so that the grayscale rise of the strong reflection neighborhood is reduced in a controlled manner, while maintaining the gradual transition along the normal without being destroyed. Since the correction process is always constrained by the upper bound of the contamination contribution, it can reduce the risk of overcompensation causing the real texture to be accidentally shaved, and make the compensation between different boundary points easier to connect in space.

[0082] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A high-definition imaging method for the bone groove region of a cochlear implant, characterized in that, The method includes: Acquire raw imaging data of the alveolar region and the acquisition parameter information corresponding to the raw imaging data; Based on the original imaging data, the bony strong reflective regions in the alveolar region are identified to obtain a bone reflective source map, wherein the bone reflective source map at least characterizes the spatial distribution of the bony strong reflective regions and the boundary information of the bony strong reflective regions. Based on the bone reflex source map, local geometric features of the bony strong reflex region are extracted, and the local geometric features include at least the region normal direction and the region curvature; Based on the local geometric features, the relative distance information of the position to be processed in the original imaging data with respect to the bony strong reflective area is determined. The bone reflection contamination propagation kernel is constructed in combination with the acquired parameter information. The position to be processed represents the pixel position in the original imaging data that needs to be evaluated for bone reflection contamination. Based on the bone reflectance source map and the bone reflectance contamination propagation kernel, digital blanking compensation is performed on the original imaging data to obtain the compensated imaging result.

2. The high-definition imaging method for the implant bone groove region based on cochlear implant according to claim 1, characterized in that, The identification methods based on the aforementioned bony strong reflex areas include: The original imaging data is preprocessed to obtain preprocessed imaging data. The preprocessing includes at least one of noise reduction processing and grayscale normalization processing. High-brightness candidate regions are extracted based on the preprocessed imaging data to obtain strong bony reflection candidate regions. Boundary detection is then performed on the strong bony reflection candidate regions to extract the candidate region boundary information. Based on the connectivity features of the candidate regions for strong bony reflexes and the boundary information of the candidate regions, the candidate regions for strong bony reflexes are screened to determine the strong bony reflex regions.

3. The high-definition imaging method for the implant bone groove region based on cochlear implant according to claim 1, characterized in that, The methods for extracting the local geometric features include: Boundary point sets are extracted from the bony strong reflection regions in the bone reflection source map to obtain the boundary point sequence of the bony strong reflection regions; Select a target boundary point from the boundary point sequence, and determine a set of boundary neighborhood points with the target boundary point as the center of a preset neighborhood range; Based on the set of boundary neighborhood points, curve fitting is performed on the local boundary of the bony strong reflex region to obtain the tangential direction at the target boundary point, and the regional normal direction at the target boundary point is determined according to the tangential direction. The local curvature at the target boundary point is determined based on the curve fitting results, wherein the local curvature is characterized by the reciprocal of the radius of curvature formed by the set of boundary neighborhood points. The region's normal direction and curvature are summed at multiple target boundary points of the bony strong reflective region to obtain the local geometric features of the bony strong reflective region.

4. The high-definition imaging method for the implant bone groove region based on cochlear implant according to claim 3, characterized in that, The target boundary points are selected as follows: the boundary point sequence is sampled at equal intervals according to a preset sampling interval to obtain an initial sampled boundary point set; for each initial sampled boundary point in the initial sampled boundary point set, the boundary reflection intensity value of the initial sampled boundary point is extracted from the original imaging data, the boundary reflection intensity value is determined at least by the pixel grayscale value at the initial sampled boundary point and the proportion of bright pixels in its neighborhood; the initial sampled boundary point set is subjected to intensity-weighted filtering based on the boundary reflection intensity value, so as to select the initial sampled boundary point whose boundary reflection intensity value meets the preset intensity condition as the target boundary point, and the boundary reflection intensity value is used as the kernel intensity parameter of the corresponding target boundary point.

5. The high-definition imaging method for the implant bone groove region based on cochlear implant according to claim 3, characterized in that, The methods for determining the relative distance information include: For each target boundary point in the bony strong reflective region, the outer normal direction is determined based on the regional normal direction at the target boundary point, and a discrete sampling chain is generated along the outer normal direction starting from the target boundary point. The discrete sampling chain consists of multiple candidate pixel positions obtained by expanding outward point by point according to a preset sampling step size. In the discrete sampling chain, at least one target pixel position corresponding to the position to be processed is determined based on a preset point selection rule. The preset point selection rule includes at least the inflection point determination of the grayscale attenuation index, the peak value determination of the gradient magnitude under the condition that the gradient direction points to the target boundary point, and the valley value determination of the local contrast compression index. Establish a mapping relationship between the target pixel position and the corresponding target boundary point, and use the corresponding target boundary point as the distance reference point of the target pixel position; Based on the projection interval between the target pixel position and the corresponding target boundary point in the outer normal direction, the relative distance information of the target pixel position relative to the bony strong reflective region is determined, wherein the projection interval is characterized by the preset sampling step size and the index of the target pixel position in the discrete sampling chain.

6. The high-definition imaging method for the implant bone groove region based on cochlear implant according to claim 1, characterized in that, A bone reflex contamination transmission kernel is constructed by combining the collected parameter information, including: Based on the acquired parameter information, a set of nuclear modulation parameters is extracted to characterize the degree of strong reflection response expansion. The set of nuclear modulation parameters includes gain parameters, exposure parameters, emission energy parameters, and scanning angle parameters. Based on the local geometric features, the directional morphological parameters and main diffusion direction of the bone reflection pollution propagation nucleus are determined, wherein the directional morphological parameters include diffusion scale parameters along the normal direction of the region and diffusion scale parameters along the tangential direction of the region. Based on the relative distance information, the distance attenuation parameter of the bone reflection contamination propagation core is determined; The directional morphological parameters and the distance attenuation parameters are adjusted according to the nuclear modulation parameter set to generate a bone reflection contamination propagation nucleus.

7. The high-definition imaging method for the implant bone groove region based on cochlear implant according to claim 6, characterized in that, Adjusting the directional shape parameter and the distance attenuation parameter includes: A kernel intensity adjustment coefficient is determined based on the gain parameter and the exposure parameter, and the diffusion scale parameter along the normal direction of the region and the diffusion scale parameter along the tangential direction of the region are synchronously amplified or reduced according to the kernel intensity adjustment coefficient. A directional offset angle is determined based on the scanning angle parameters, and the main diffusion direction of the bone reflection contamination propagation nucleus is rotated and corrected according to the directional offset angle. An attenuation adjustment coefficient is determined based on the emitted energy parameters, and the distance attenuation parameters are adjusted according to the attenuation adjustment coefficient. Bone reflection contamination propagation nuclei are generated based on the adjusted diffusion scale parameters along the regional normal direction, the adjusted diffusion scale parameters along the regional tangential direction, the adjusted main diffusion direction, and the adjusted distance attenuation parameters.

8. The high-definition imaging method for the implant bone groove region based on cochlear implant according to claim 7, characterized in that, The bone reflection pollution propagation kernel is characterized by a set of kernel parameters that are associated one-to-one with each target boundary point of the strong bone reflection region. The set of kernel parameters includes kernel intensity parameters, diffusion scale parameters along the normal direction, diffusion scale parameters along the tangential direction, main diffusion direction parameters, and distance attenuation parameters. Based on the set of kernel parameters, a kernel weight function is determined in a local coordinate system with the corresponding target boundary point as the origin and formed by its tangential and normal directions.

9. The high-definition imaging method for the implant bone groove region based on cochlear implant according to claim 1, characterized in that, The compensation methods for the digital blanking compensation include: Based on the bone reflectance source map and the bone reflectance contamination propagation kernel, a contamination contribution field of the outer neighborhood is generated for each target boundary point of the strong bony reflectance region in the corresponding local coordinate system to obtain the upper bound information of the contamination contribution of each outer neighborhood position. For the location to be processed, the limited compensation amount for the location to be processed is determined based on the relative distance information relative to the corresponding target boundary point and the pollution contribution upper bound information. A normal sampling chain consistency constraint is constructed based on the outer normal direction associated with the target boundary point, and the constrained compensation amount is corrected for normal consistency based on the normal sampling chain consistency constraint. The corresponding unprocessed position is compensated based on the limited compensation amount after normal consistency correction.

10. An endoscopic imaging cochlea, used to realize the high-definition imaging method of the implant bone alveolar region based on a cochlear implant as described in any one of claims 1-9, characterized in that, The endoscopic imaging cochlea includes: A cochlear mold, the cochlear mold including an insertion part and a probe end disposed at the distal end of the insertion part, the probe end being used to enter the cochlear cavity and perform near-field imaging of the alveolar region; An imaging component is disposed at the end of the probe. The imaging component includes an imaging sensor and an imaging optics corresponding to the optical axis of the imaging sensor, and is used to acquire raw imaging data of the alveolar region. An illumination component, disposed at the end of the probe and facing the framing direction, is used to provide illumination light to the alveolar region to support imaging by the imaging component; A processing component, electrically connected to the imaging component, is configured to generate a bone reflectance source map based on the original imaging data, extract local geometric features of the bony strong reflectance area, determine the relative distance information of the location to be processed relative to the bony strong reflectance area, construct a bone reflectance contamination propagation kernel by combining the acquired parameter information, and perform digital blanking compensation on the original imaging data based on the bone reflectance source map and the bone reflectance contamination propagation kernel to output the imaging result. An output component, electrically connected to the processing component, is used to output the compensated imaging result to a display terminal.

Images