Probe angle calibration method suitable for flexible ureteroscope tail end

Abstract

The invention discloses a probe lens angle calibration method suitable for the tail end of a flexible ureteroscope, and relates to the technical field of probe lens angle calibration, and the method comprises the following steps: presetting an initial adjustment time as the basic duration of each angle adjustment before the flexible ureteroscope is subjected to ureteral examination; in the angle adjusting process, ureter image data in a target detection area are collected in real time through an imaging unit arranged in the flexible lens. According to the method, an intelligent structural complexity evaluation mechanism based on the convolutional neural network is introduced, so that the system can dynamically identify the anatomical complexity of a target area, and adaptively adjust the angle adjustment duration according to an evaluation result, thereby overcoming the technical bottleneck that a target cannot be accurately aligned in a high-tortuosity area in the conventional fixed adjustment time, and improving the accuracy of the target anatomy. It is ensured that the probe of the soft lens can be adjusted more carefully and stably in a complex area, the problems of image blurring, distortion or target deviation and the like are reduced, and therefore the risks of missed diagnosis and misdiagnosis are greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of scope angle calibration, and in particular to a scope angle calibration method applicable to the distal end of a flexible ureteroscope. Background Art

[0002] Ureteroscope end-probe angle calibration refers to the precise adjustment of the angle of the flexible end-probe during urinary tract examinations to ensure it is accurately aligned with the target area, providing the best field of view and image. This technology is crucial for accurately assessing the location of lesions or stones within the ureter, avoiding incomplete examinations or misdiagnoses due to probe angle deviations. Through this calibration, the flexible endoscope can be flexibly positioned in the complex urinary tract environment, ensuring that doctors can perform accurate diagnosis and treatment operations, thereby improving the effectiveness and safety of treatment.

[0003] In the prior art, urinary tract examinations usually reserve a fixed adjustment time for the angle adjustment of the flexible endoscope probe, in order to ensure that the probe can be stably and accurately aimed at the target area. The fixed adjustment time design can complete the adjustment of the probe angle within a certain time range, thereby avoiding image distortion, blurring or instability caused by too fast or too slow adjustment. The purpose of this method is to optimize image quality and ensure that doctors can accurately observe the structure of the urinary tract and potential lesions, such as stones, tumors, etc., and make effective diagnosis and treatment based on this. The fixed adjustment time can meet the needs of urinary tract examinations under normal circumstances, and improve the accuracy of diagnosis by reducing errors caused by excessive or unstable adjustment.

[0004] However, the existing technology has the following problems: The ureteral anatomy varies significantly between patients, and examinations can involve complex tortuosity, particularly in lesioned areas. These complex structural variations require more flexible and dynamic angle adjustment of the flexible endoscope probe, but fixed adjustment times cannot respond to these changes in real time. Fixed adjustment times fail to fully account for the dynamic changes in the ureteral tortuosity during angle adjustment, potentially leading to insufficient angle adjustment precision and decreased image quality, including blurring, distortion, or distortion. This reduced image quality directly impacts the clarity of critical areas such as lesions, stones, or tumors, which in turn can affect physicians' interpretation of examination results, increase the risk of missed or misdiagnosis, and lead to incorrect treatment plans. As a result, patients may miss optimal treatment opportunities, their condition may worsen, metastasize, or worsen, and in severe cases, even be life-threatening. Therefore, existing fixed adjustment time designs cannot effectively adapt to the dynamic changes in the complex ureteral anatomy, posing a potential risk of compromising diagnostic accuracy and treatment effectiveness.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a probe angle calibration method suitable for the end of a flexible ureteroscope. By introducing an intelligent structural complexity evaluation mechanism based on a convolutional neural network, the system can dynamically identify the anatomical complexity of the target area and adaptively adjust the angle adjustment time according to the evaluation results, thereby overcoming the technical bottleneck that the traditional fixed adjustment time cannot accurately align the target in highly tortuous areas. Especially in the area of ​​ureteral lesions, after adopting this solution, the flexible endoscope can achieve more detailed and smooth probe adjustment in complex areas, reduce the occurrence of problems such as image blur, distortion or target deviation, thereby greatly reducing the risk of missed diagnosis and misdiagnosis, and solving the problems in the above-mentioned background technology.

[0007] To achieve the above-mentioned object, the present invention provides the following technical solution: a method for calibrating the probe angle of a flexible ureteroscope, comprising the following steps: Before performing a ureteral examination with a flexible endoscope, an initial adjustment time is preset as the basic duration for each angle adjustment to ensure fast and accurate angle adjustment within the conventional area, maximizing image stability, visual angle continuity, and operational consistency. During the angle adjustment process, the imaging unit built into the soft endoscope collects ureteral image data within the target detection area in real time. The collected raw image data is input into the image processing unit for preprocessing operations to improve image quality and structural feature significance. This provides more reliable and stable input data for subsequent feature extraction, making the neural network evaluation more accurate and robust. For the pre-processed ureteral image data, the core structural features reflecting the ureteral tortuosity are extracted. After processing the extracted features using feature engineering technology, the tortuosity complexity of the ureteral internal structure is quantified. The extracted features are input as feature vectors into the trained convolutional neural network model, which automatically analyzes the nonlinear feature combination relationship in the image and evaluates the structural complexity of the target detection area. According to the output results of the convolutional neural network model, the target detection area currently observed by the probe is divided into two categories, including regular areas and high-complexity tortuous areas; For target detection areas divided into regular areas, the initial adjustment time is kept unchanged to ensure that the angle adjustment can be completed quickly and accurately. For target detection areas divided into high-complexity tortuous areas, the adjustment time extension mechanism is triggered, and the initial adjustment time is adaptively extended based on the structural complexity evaluation results of the target detection area by the convolutional neural network. The actual adjustment time after extension is used as the control time for the current angle adjustment, which is used to drive the probe to perform slow and precise angle correction.

[0008] Preferably, the initial adjustment time is set through clinical experience and routine operation optimization, and the value range is between 2-3 seconds.

[0009] Preferably, the specific steps of collecting real-time image data of the interior of the ureter by the built-in imaging unit of the soft endoscope are as follows: When the soft endoscope operation begins, the built-in imaging unit of the soft endoscope is started, and the imaging unit automatically adjusts the brightness of the light source according to the current lighting conditions to ensure that a clear image of the inside of the ureter is captured; The collected image data is transmitted to the processing unit of the soft endoscope in real time through the transmission line to decode and format the image data; The processed image data is displayed on the display screen in real time, allowing the doctor to observe the ureteral area currently observed by the soft endoscope probe.

[0010] Preferably, for the preprocessed ureteral image data, core structural features reflecting the tortuosity of the ureter are extracted therefrom, wherein the extracted features include the degree of rotation of the ureter in its axial direction and the structural density of the ureter in the local area. After the extracted features are processed using feature engineering technology, an axial torsion factor and a local density factor are generated respectively, and the tortuosity complexity of the internal structure of the ureter is quantified by the axial torsion factor and the local density factor.

[0011] Preferably, the extracted axial torsion factor and local density factor are input as feature vectors into the trained convolutional neural network model, and the convolutional neural network model automatically analyzes the combination relationship of the axial torsion factor and the local density factor to generate a topological complexity coefficient for characterizing the structural complexity, and the structural complexity of the target detection area is quantitatively evaluated based on the topological complexity coefficient.

[0012] Preferably, after obtaining the topological complexity coefficient generated when the convolutional neural network model evaluates the structural complexity of the target detection area, the topological complexity coefficient is compared and analyzed with a pre-set topological complexity coefficient reference threshold to divide the target detection area. The specific division steps are as follows: If the topological complexity coefficient is greater than the topological complexity coefficient reference threshold, the target detection area currently observed by the probe is divided into a high-complexity tortuous area; If the topology complexity coefficient is less than or equal to the topology complexity coefficient reference threshold, the target detection area currently observed by the probe is divided into a regular area.

[0013] Preferably, for the target detection area divided into regular areas, the initial adjustment time is kept unchanged to ensure that the angle adjustment of the soft endoscope probe can be completed quickly and accurately. In this case, the control time is equal to the preset initial adjustment time, that is: , where: is the actual adjustment time for the general area (keep the same as the initial adjustment time), The preset initial adjustment time is usually 2 to 3 seconds, which is optimized through clinical experience; For target detection areas classified as high-complexity tortuous areas, the adjustment time extension mechanism is triggered. The initial adjustment time is adaptively extended based on the convolutional neural network evaluation results. The extended adjustment time is dynamically adjusted based on the relationship between the topological complexity coefficient and the topological complexity coefficient reference threshold. The specific formula is as follows: , where: is the actual adjustment time of the high-complexity tortuous area, after adaptive extension, is the topological complexity coefficient of the target detection area calculated by the convolutional neural network, which indicates the structural complexity of the target detection area. is the reference threshold of topological complexity, which is used as a standard value to adjust the extension ratio. k is the adjustment coefficient, which indicates the influence of the topological complexity coefficient on the extension of the adjustment time. .

[0014] Preferably, for the target detection area, the specific steps of using feature engineering technology to process the degree of rotation of the ureter in its axial direction to generate an axial torsion factor are as follows: First, the local rotation angle of each segment is extracted from the ureter image of the target detection area by image processing technology. Specifically, the rotation angle is calculated by tracing the center line of the ureter or the tangent direction change along the axis of the ureter. To this end, two adjacent points are set and The vector between ,in: , where: 、 、 、 Represents the coordinate point. By calculating the angle change between adjacent vectors, the local rotation angle is obtained. The calculation expression of the local rotation angle is: , where: is the local rotation angle of the i-th segment, reflecting the degree of rotation of each segment, Represents a vector and vector The dot product of and Represents vectors and vector The model; Next, according to the local rotation angle of each segment Calculate the axial torsion factor, which measures the degree of axial torsion by the weighted cumulative value of the local rotation angles and takes into account the effect of local tortuosity on the total rotation complexity. The axial torsion factor is calculated using the following formula: , where: is the axial torsion factor of the target detection area, which represents the overall complexity of the rotation within the target detection area. are two adjacent points in the i-th segment and The axial distance between them reflects the length of each segment. is a positive adjustment parameter used to balance the relationship between the local rotation angle and the axial distance, and n is the total number of segments considered.

[0015] Preferably, for the target detection area, the specific steps of using feature engineering technology to process the structural density of the ureter in the local area to generate a local density factor are as follows: Perform image enhancement and binarization on the target detection image obtained by the soft mirror imaging unit to obtain a binary structure image with clear structural boundaries , where a pixel value of 1 indicates the presence of ureteral tissue structure and 0 indicates the absence of structure. Then, a local sliding window with a radius of r is defined with any reference point in the target detection area as the center. , and based on the local sliding window The distribution of internal structure pixels is used to calculate the structure occupancy rate of the local sliding window area. The calculation formula is as follows: , where: is the total number of pixels in the local sliding window; The structure occupancy rate in the local sliding window area represents the degree of structure filling within the local sliding window. The higher the value, the denser the structure in the local area, indicating that the target detection area may have multiple bends, folds or spatial distortions. In order to further improve the recognition ability of complex and tortuous areas, the first scale radius is set With the second scale radius (in ), and obtain the first scale radius respectively With the second scale radius The structural occupancy of and Based on this, a multi-scale disturbance response model is constructed, and a local density factor is generated to quantify the local tortuosity complexity. The generation expression of the local density factor is as follows: , where: is the local density factor, and are nonlinear amplification factors, respectively, used to enhance the sensitivity to the local structure absolute density and scale perturbation ratio. The local density factor, based on the absolute structure density, introduces the cross-scale perturbation variation ratio to capture the high-frequency structural changes of the ureter in a small space.

[0016] In the above technical solution, the technical effects and advantages provided by the present invention are: By introducing an intelligent structural complexity assessment mechanism based on a convolutional neural network, this invention enables the system to dynamically identify the anatomical complexity of the target area and adaptively adjust the angle adjustment duration based on the assessment results, thereby overcoming the technical bottleneck of the traditional fixed adjustment time that cannot accurately align the target in highly tortuous areas. Especially in the area of ​​ureteral lesions, after adopting this solution, the soft endoscope can achieve more detailed and smooth probe adjustment in complex areas, reducing the occurrence of problems such as image blur, distortion, or target deviation, thereby significantly reducing the risk of missed diagnosis and misdiagnosis, improving the visibility and safety of doctors' operations, and ultimately providing patients with more timely, accurate, and effective clinical intervention support. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0018] Figure 1 The present invention is a flowchart of a method for calibrating the angle of a probe at the end of a flexible ureteroscope. DETAILED DESCRIPTION

[0019] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0020] The present invention provides Figure 1 A method for calibrating the angle of a probe at the distal end of a flexible ureteroscope is shown, comprising the following steps: Before performing a ureteral examination with a flexible endoscope, an initial adjustment time is preset as the basic duration for each angle adjustment to ensure fast and accurate angle adjustment within the conventional area, maximizing image stability, visual angle continuity, and operational consistency. This initial adjustment time (e.g., 2-3 seconds) is optimized through clinical experience and routine operation and is suitable for most ureteral regions with relatively straight anatomical structures and minimal curvature (typically referring to these ureteral regions with relatively straight anatomical structures and minimal curvature). Each time the flexible endoscope changes angle, the control system flexibly adjusts according to this time, allowing the distal probe to slowly and steadily align with the target area.

[0021] Optimizing settings through clinical experience and routine operation means that in clinical practice, physicians and technicians gradually determine an initial adjustment time that is suitable for most patients based on extensive practical experience, historical data, and observations of different patient anatomy. This setting (e.g., 2-3 seconds) is based on repeated trials and adjustments. It ensures that, under routine conditions, the flexible endoscope probe can accurately adjust the angle within the common ureteral anatomy without delay and effectively aligns with the target area. This process not only considers operational efficiency but also the precision of probe angle adjustment, aiming to meet the needs of actual clinical practice and ensure optimal image quality and diagnostic results.

[0022] During the angle adjustment process, the imaging unit built into the soft endoscope collects ureteral image data within the target detection area in real time. The collected raw image data is input into the image processing unit for preprocessing operations to improve image quality and structural feature significance. This provides more reliable and stable input data for subsequent feature extraction, making the neural network evaluation more accurate and robust. The specific steps for real-time acquisition of internal ureteral image data using the built-in imaging unit of the soft endoscope are as follows: Step 1: Start the imaging device and initialize image acquisition; When a soft endoscope is operated, the imaging unit (typically a CMOS or CCD sensor) built into the endoscope is activated. The system automatically adjusts the light source brightness based on the current lighting conditions to ensure a sufficiently clear image. At this point, the imaging unit begins capturing real-time image data of the area where the soft endoscope probe is located, ensuring that each frame accurately reflects the current field of view.

[0023] Step 2: Transmit image data to the processing system; The collected image data is transmitted in real time to the processing unit of the soft mirror through transmission lines such as optical fibers or cables. The processing system decodes and formats the transmitted image data for subsequent analysis and processing. At this time, the imaging system maintains a high frame rate (usually 30 fps or higher), ensuring smooth and uninterrupted image flow that continuously reflects the details of the ureter.

[0024] Step three: real-time display and feedback; The processed image data is displayed in real time on the display screen, and the doctor can clearly see the ureter region observed by the current soft mirror probe. The display screen not only presents the current image, but also optimizes the image brightness, contrast, and other image parameters as needed, adjusts the probe angle in time, and ensures the best view in different lighting or complex areas. At the same time, the system will continuously feedback to the doctor according to the real-time image updates, assisting them in making diagnosis decisions.

[0025] The internal image data of the ureter usually includes information such as mucosal profile, blood vessel direction, structural deformation, brightness distribution, etc. within the field of view. Real-time image acquisition provides high-fidelity input for subsequent analysis and is the basis for identifying complex anatomical structures. The non-lagged image flow can be synchronized with the adjustment action, improving the real-time performance and accuracy of the system response.

[0026] Preprocessing operations include image denoising, grayscale normalization, contrast enhancement, edge extraction, and light correction. Especially in non-ideal imaging conditions (such as light reflection and mirror blur), the preprocessing process can significantly improve the clarity and structural recognizability of the image. In addition, time-dimension filtering processing is also performed on consecutive frames of images, which can extract structural change trends and capture dynamic bending characteristics. For the preprocessed ureter image data, the core structural features reflecting the tortuosity of the ureter are extracted, and the extracted features are processed using feature engineering techniques to quantify the tortuosity complexity of the internal structure of the ureter. For the preprocessed ureter image data, the core structural features reflecting the tortuosity of the ureter are extracted, and the extracted features include the degree of rotation of the ureter in its axial direction and the structural density of the ureter in the local area. The extracted features are processed using feature engineering techniques to generate an axial torsion factor and a local density factor, respectively, and the axial torsion factor and the local density factor are used to quantify the tortuosity complexity of the internal structure of the ureter. For the target detection area currently observed by the probe, if the degree of axial rotation of the ureter in the target detection area is high, it usually indicates that the tortuosity complexity of the ureter in this area is high. The reason is that the axial rotation of the ureter represents its spatial distortion or rotation in three-dimensional space. Axial rotation usually occurs when the ureter is strongly bent, torsion or narrowed. A larger axial rotation angle indicates that the ureter has undergone more complex deformation in this area, which may be due to structural problems (such as stones, tumors or inflammation) that cause changes in the ureter's morphology. Strong rotation is usually accompanied by multiple tortuosity points or extremely curved areas, which makes the anatomical structure of the area more complex and increases the difficulty of treatment and diagnosis. Therefore, a high degree of axial rotation is usually directly related to structural changes with high tortuosity, which can reflect the high complexity of the ureter in this area.

[0027] For the target detection area, the specific steps of using feature engineering technology to process the degree of ureteral rotation on its axis and generate the axial torsion factor are as follows: First, the local rotation angle of each segment is extracted from the ureter image of the target detection area by image processing technology. Specifically, the rotation angle is calculated by tracing the center line of the ureter or the tangent direction change along the axis of the ureter. To this end, two adjacent points are set and The vector between ,in: , where: 、 、 、 Represents the coordinate point. By calculating the angle change between adjacent vectors, the local rotation angle is obtained. The calculation expression of the local rotation angle is: , where: is the local rotation angle of the i-th segment, reflecting the degree of rotation of each segment, Represents a vector and vector The dot product of and Represents vectors and vector The model; The purpose of this step is to quantify the degree of ureteral rotation at each small segment, identify axial rotation, and provide baseline data for subsequent evaluation.

[0028] Next, according to the local rotation angle of each segment Calculate the axial torsion factor, which measures the degree of axial torsion by the weighted cumulative value of the local rotation angles and takes into account the effect of local tortuosity on the total rotation complexity. The axial torsion factor is calculated using the following formula: , where: is the axial torsion factor of the target detection area, which represents the overall complexity of the rotation within the target detection area. are two adjacent points in the i-th segment and The axial distance between them reflects the length of each segment. is a positive adjustment parameter used to balance the relationship between the local rotation angle and the axial distance, and n is the total number of segments considered; This step is designed to ensure that regions with larger rotation angles contribute more to the axial torsion factor, and that shorter regions (i.e., densely tortuous regions) have a higher weight in the axial torsion factor. Through this step, the axial torsion complexity of the entire ureteral target region can be effectively assessed.

[0029] The axial torsion factor, generated by processing the degree of ureteral rotation along its axis using feature engineering techniques, indicates that for the target detection region, a higher performance value of the axial torsion factor indicates greater tortuosity complexity within the target detection region, while a lower performance value indicates less tortuosity complexity within the target detection region. This is because the axial torsion factor assesses structural complexity by quantifying the degree of rotation and local deformation of the ureter within the target region. Specifically, when the ureter undergoes significant rotation or twisting within a certain region, it indicates significant curvature or stenosis, leading to a high degree of anatomical complexity. The axial torsion factor converts this degree of rotation and deformation into a numerical value by weighted accumulation of local rotation angles. A higher axial torsion factor reflects more dramatic structural changes and greater tortuosity in the region, while a lower axial torsion factor indicates a relatively simple anatomical structure with less curvature.

[0030] When the structural density of the ureter in the target detection area currently observed by the probe is high in a local area, it usually indicates that the tortuosity complexity of this area is high. The reason is that in the image, the structural density reflects the frequency of path changes and the degree of spatial occupancy of the ureter in a limited space. When the ureter frequently turns, folds or twists in a certain area, the contour lines presented in the image will be denser, and the distance between the curves will become smaller, forming a local overlap or high proximity. This dense layout often means that the ureteral structure in this area has continuous bending or complex morphological changes in multiple directions, which corresponds to an anatomical morphology with high tortuosity. Therefore, local structural density is one of the important image features for evaluating the complexity of ureteral tortuosity and has a strong spatial distribution indication.

[0031] For the target detection area, the specific steps of using feature engineering technology to process the structural density of the ureter in the local area to generate the local density factor are as follows: Perform image enhancement and binarization on the target detection image obtained by the soft mirror imaging unit to obtain a binary structure image with clear structural boundaries , where a pixel value of 1 indicates the presence of ureteral tissue structure and 0 indicates the absence of structure. Then, a local sliding window with a radius of r is defined with any reference point in the target detection area as the center. , and based on the local sliding window The distribution of internal structure pixels is used to calculate the structure occupancy rate of the local sliding window area. The calculation formula is as follows: , where: is the total number of pixels in the local sliding window; The structure occupancy rate in the local sliding window area represents the degree of structure filling within the local sliding window. The higher the value, the denser the structure in the local area, indicating that the target detection area may have multiple bends, folds or spatial distortions. This step traverses the target detection image at high spatial resolution to capture the local spatial filling of the ureteral structure around each location. Highly filled areas indicate multiple bends or superpositions, which is a warning indicator of high tortuosity complexity.

[0032] In order to further improve the recognition ability of complex and tortuous areas, the first scale radius is set With the second scale radius (in ), and obtain the first scale radius respectively With the second scale radius The structural occupancy of and Based on this, a multi-scale disturbance response model is constructed, and a local density factor is generated to quantify the local tortuosity complexity. The generation expression of the local density factor is as follows: , where: is the local density factor, and are nonlinear amplification factors, respectively, used to enhance the sensitivity of the response to the local structure absolute density and scale perturbation ratio. The local density factor, based on the consideration of the absolute structure density, introduces the cross-scale perturbation change ratio to capture the high-frequency structural changes of the ureter in a small space; By integrating the relative disturbance relationship of structural occupancy at different scales, a dynamic response model for complex structures in local space is established, which can highly sensitively identify morphological features such as structural reentry and concentrated bending, thereby achieving accurate identification of highly tortuous and complex areas of the ureter.

[0033] The local density factor indicates that, for the target detection region, a larger performance value of the local density factor, generated by feature engineering techniques to process the structural density of the ureter within the local region, indicates a higher tortuosity complexity within the target detection region; conversely, a lower performance value indicates a lower tortuosity complexity within the target detection region. This is because the local density factor, as an image feature parameter reflecting the local structural complexity of the ureter, has a positive correlation with the tortuosity complexity of the ureter within the target detection region. Specifically, a larger local density factor indicates a high degree of spatial filling, frequent path bending, and complex morphologies with multiple turns, twists, or overlaps within the local region. This dense structural arrangement directly reflects a high tortuosity complexity. A smaller local density factor indicates a sparse distribution of ureteral structures within the region, a relatively flat path, and low spatial variation, corresponding to a low tortuosity complexity. Therefore, as a quantitative indicator of structural complexity generated through feature engineering, a higher local density factor value more significantly reveals the high tortuosity characteristics of the ureter within the target detection region and serves as an important basis for determining the degree of local anatomical complexity.

[0034] The extracted features are input as feature vectors into the trained convolutional neural network model, which automatically analyzes the nonlinear feature combination relationship in the image and evaluates the structural complexity of the target detection area. The extracted axial torsion factor and local density factor are input as feature vectors into the trained convolutional neural network model. The convolutional neural network model automatically analyzes the combination relationship of the axial torsion factor and the local density factor, generates a topological complexity coefficient for characterizing structural complexity, and quantitatively evaluates the structural complexity of the target detection area based on the topological complexity coefficient.

[0035] A trained convolutional neural network model refers to a neural network system built on a deep learning architecture that has completed the training phase. Its core goal is to intelligently assess structural complexity by performing multi-level feature extraction, combination, and classification on input images or feature vectors. In this scenario, the convolutional neural network (CNN) is no longer used as a model for traditional image classification, but is instead applied to a structural complexity recognition task. Specifically, it performs joint modeling and in-depth analysis of the "axial torsion factor" and "local density factor" extracted during soft endoscopy.

[0036] "Training complete" means that the CNN model has undergone a systematic supervised learning process. First, developers construct a labeled dataset based on a large number of ureteral image data acquired through flexible endoscopy. From this dataset, they extract image feature parameters related to ureteral structural complexity, such as the axial torsion factor (reflecting the degree of spatial twisting of the ureter) and the local density factor (quantifying the density of local structures). These features, along with corresponding structural complexity labels (e.g., low complexity, medium complexity, and high complexity), are then fed into the neural network model as training samples. Leveraging its powerful multi-layer convolutional architecture, convolutional neural networks automatically learn the nonlinear relationship between these two features, iterating through updates to network parameters (such as convolution kernel weights, bias terms, and activation function outputs), ultimately enabling the model to discriminate the structural complexity of input feature vectors. Training is considered complete when the model achieves high classification accuracy on the validation set and the error converges steadily. At this point, the model has developed a mature parameter set capable of performing inference and prediction tasks on unseen examples in real-world applications.

[0037] In this system, the significance of using the trained CNN model to jointly analyze the "axial torsion factor" and the "local density factor" is that it can not only identify the impact of structural changes brought about by each individual indicator, but more importantly, it can automatically learn the deep correlation pattern between the two indicators from the data. For example, even if the degree of axial distortion in some areas is not high, due to the abnormally high density, higher structural complexity may be formed; while other areas may have normal density but if accompanied by nonlinear distortion or local rotation, they will also constitute inspection difficulties. This kind of nonlinear, cross-interaction relationship pattern is often not able to be captured by traditional artificial rules or linear regression modeling. By stacking multiple convolutional layers, nonlinear activation layers and pooling layers, convolutional neural networks can construct a complex feature expression system, integrating these difficult-to-quantify interaction factors into a compact "topological complexity coefficient".

[0038] The "topological complexity coefficient" is not a simple numerical weighting or linear combination, but a comprehensive indicator after nonlinear projection in a high-dimensional space, which can accurately quantify the anatomical complexity of the target area. In addition, the trained CNN model also has a certain generalization ability. That is, it can not only identify typical complex structures that appear in the training set, but also give reasonable complexity judgments for variant, rare, and even pathological structures encountered in clinical practice. This is because the model improves its adaptability to different types of structural combination patterns through data augmentation, regularization, and the introduction of multiple samples during the training phase. Therefore, it can be widely applied to image structure recognition in different patients, with different ureteral course characteristics, and under different operating environments.

[0039] In summary, the "trained convolutional neural network model" described in this article is essentially an intelligent assessment system with "autonomous learning," "nonlinear analysis," and "structural perception capabilities." It is trained based on real clinical data and incorporates deep feature representations of multiple structural parameters. Through the model's solidified multi-layer neural network architecture, it quantitatively assesses the structural complexity of ureteral images and generates a "topological complexity coefficient," providing a scientific basis for subsequent image enhancement processing, soft endoscope control strategies, and clinical intervention decisions. This approach fully leverages the advantages of artificial intelligence models in medical image feature extraction, nonlinear reasoning, and real-time feedback, addressing the shortcomings of traditional static adjustment strategies in identifying individual anatomical differences and dynamic structures.

[0040] The convolutional neural network model is not specifically limited here, and can achieve the axial torsion factor and the local density factor Perform comprehensive analysis to generate topological complexity coefficients In order to realize the technical solution of the present invention, the present invention provides a specific implementation method; the topological complexity coefficient The generated expression is: , where 、 Axial torsion factor and the local density factor The preset scaling factor of 、 The preset proportional coefficient refers to the axial torsion factor of the two characteristic indicators when calculating the topological complexity coefficient. and the local density factor The artificially set weight parameters used to weigh the influence of the two in the final comprehensive evaluation are respectively denoted as and These proportional coefficients are not obtained through real-time learning or automatic calculation, but are constants set in advance based on experimental experience, actual scenarios or expert knowledge when designing the model. Specifically, and The role of is to regulate the contribution of the two core indicators to the overall complexity evaluation in the process of generating the topological complexity coefficient. For example, when Greater than When axial torsion is large, the model places greater emphasis on the impact of structural complexity; conversely, it emphasizes the role of local density. This pre-set design allows the model to adapt specifically to the structural characteristics of different types of ureters, achieving flexibility and precision control of complexity judgment in practical applications. Therefore, the introduction of the preset scaling coefficient not only provides physical support for the interpretability of the convolutional neural network output, but also provides a parameterized adjustment method for the system's adaptability to different complex structures.

[0041] It can be seen from the topological complexity coefficient that for the target detection area, the greater the performance value of the axial torsion factor generated after the feature engineering technology is used to process the degree of rotation of the ureter in its axial direction, the greater the performance value of the local density factor generated after the feature engineering technology is used to process the structural density of the ureter in the local area. That is, the greater the performance value of the topological complexity coefficient generated when the trained convolutional neural network model is used to evaluate the structural complexity of the target detection area, the higher the tortuosity complexity of the internal structure of the ureter in the target detection area, and vice versa.

[0042] According to the output results of the convolutional neural network model, the target detection area currently observed by the probe is divided into two categories, including regular areas and high-complexity tortuous areas; After obtaining the topological complexity coefficient generated by the convolutional neural network model when evaluating the structural complexity of the target detection area, the topological complexity coefficient is compared and analyzed with the pre-set topological complexity coefficient reference threshold to divide the target detection area. The specific division steps are as follows: If the topological complexity coefficient is greater than the topological complexity coefficient reference threshold, the target detection area currently observed by the probe is divided into a high-complexity tortuous area; If the topology complexity coefficient is less than or equal to the topology complexity coefficient reference threshold, the target detection area currently observed by the probe is divided into a regular area.

[0043] For target detection areas classified as regular areas, the initial adjustment time remains unchanged to ensure that angle adjustment can be completed quickly and accurately. For target detection areas classified as highly complex and tortuous areas, the adjustment time extension mechanism is triggered. Based on the structural complexity assessment results of the target detection area by the convolutional neural network, the initial adjustment time is adaptively extended. The actual adjustment time after the extension is used as the control time for the current angle adjustment, which is used to drive the probe to perform slow and precise angle correction. For the target detection area divided into regular areas, keep the initial adjustment time unchanged to ensure that the angle adjustment of the soft endoscope probe can be completed quickly and accurately. At this time, the control time is equal to the preset initial adjustment time, that is: , where: is the actual adjustment time for the general area (keep the same as the initial adjustment time), The preset initial adjustment time is usually 2 to 3 seconds, which is optimized through clinical experience; This step ensures that the flexible endoscope probe can be quickly adjusted to the desired angle within the normal area, ensuring efficient and accurate inspections. This is particularly useful in areas with simple structures. Maintaining the same initial adjustment time improves operational consistency and diagnostic efficiency, reducing time wastage.

[0044] For target detection areas classified as high-complexity tortuous areas, the adjustment time extension mechanism is triggered. The initial adjustment time is adaptively extended based on the convolutional neural network evaluation results. The extended adjustment time is dynamically adjusted based on the relationship between the topological complexity coefficient and the topological complexity coefficient reference threshold. The specific formula is as follows: , where: is the actual adjustment time of the high-complexity tortuous area, after adaptive extension, is the topological complexity coefficient of the target detection area calculated by the convolutional neural network, which indicates the structural complexity of the target detection area. is the reference threshold of topological complexity, which is used as a standard value to adjust the extension ratio. k is the adjustment coefficient, which indicates the influence of the topological complexity coefficient on the extension of the adjustment time. ; The key to this step is adaptively adjusting the angle adjustment time based on the topological complexity coefficient. If the target area's structural complexity is high and the topological complexity coefficient increases, the adjustment time will be extended, allowing the soft endoscope probe to make more precise and slower angle corrections, ensuring precise alignment of the target area within complex and tortuous areas. This dynamic extension mechanism effectively avoids inaccurate probe adjustments due to insufficient time and improves imaging quality in highly complex areas.

[0045] The above solution can significantly improve the angle adjustment accuracy and imaging stability of the soft endoscope in the complex and tortuous areas of the ureter, thereby enhancing the overall reliability and diagnostic accuracy of clinical examinations. This method introduces an intelligent structural complexity evaluation mechanism based on a convolutional neural network, enabling the system to dynamically identify the anatomical complexity of the target area and adaptively adjust the angle adjustment time based on the evaluation results, thereby overcoming the technical bottleneck of the traditional fixed adjustment time that cannot accurately aim at the target in highly tortuous areas. Especially in areas with ureteral lesions, such as stones, stenosis or tumors and other structural abnormalities, the sensitivity and stability of angle adjustment are crucial to imaging quality and lesion identification. After adopting this solution, the soft endoscope can achieve more detailed and smooth probe adjustment in complex areas, reducing the occurrence of problems such as image blur, distortion or target deviation, thereby significantly reducing the risk of missed diagnosis and misdiagnosis, improving the visibility and safety of doctors' operations, and ultimately providing patients with more timely, accurate and effective clinical intervention support.

[0046] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0047] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0048] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for calibrating the angle of a probe at the end of a flexible ureteroscope, characterized in that: The following steps are involved: Before performing ureteral examination with a flexible endoscope, an initial adjustment time is preset as the basic duration for each angle adjustment; During the angle adjustment process, the imaging unit built into the soft endoscope collects ureteral image data within the target detection area in real time. The collected raw image data is input into the image processing unit for pre-processing to improve image quality and structural feature significance. For the pre-processed ureteral image data, the core structural features reflecting the ureteral tortuosity are extracted. After processing the extracted features using feature engineering technology, the tortuosity complexity of the ureteral internal structure is quantified. The extracted features are input as feature vectors into the trained convolutional neural network model, which automatically analyzes the nonlinear feature combination relationship in the image and evaluates the structural complexity of the target detection area. According to the output results of the convolutional neural network model, the target detection area currently observed by the probe is divided into two categories, including regular areas and high-complexity tortuous areas; For target detection areas divided into regular areas, the initial adjustment time remains unchanged; for target detection areas divided into high-complexity tortuous areas, the adjustment time extension mechanism is triggered, and the initial adjustment time is adaptively extended based on the structural complexity evaluation results of the target detection area by the convolutional neural network. The actual adjustment time after extension is used as the control time for the current angle adjustment, which is used to drive the probe to correct the angle.

2. A method for calibrating the probe angle of a flexible ureteroscope according to claim 1, characterized in that: The initial adjustment time is set through clinical experience and routine operation optimization, and the value range is between 2-3 seconds.

3. A method for calibrating the probe angle of a flexible ureteroscope according to claim 1, characterized in that: The specific steps for real-time acquisition of internal ureteral image data using the built-in imaging unit of the soft endoscope are as follows: When the soft endoscope operation begins, the built-in imaging unit of the soft endoscope is started, and the imaging unit automatically adjusts the brightness of the light source according to the current lighting conditions to ensure that a clear image of the inside of the ureter is captured; The collected image data is transmitted to the processing unit of the soft endoscope in real time through the transmission line to decode and format the image data; The processed image data is displayed on the display screen in real time, allowing the doctor to observe the ureteral area currently observed by the soft endoscope probe.

4. A method for calibrating the probe angle of a flexible ureteroscope according to claim 1, characterized in that: For the preprocessed ureteral image data, core structural features reflecting the tortuosity of the ureter are extracted. The extracted features include the degree of rotation of the ureter in its axial direction and the structural density of the ureter in the local area. After processing the extracted features using feature engineering technology, the axial torsion factor and local density factor are generated respectively. The axial torsion factor and local density factor are used to quantify the tortuosity complexity of the internal structure of the ureter.

5. A method for calibrating the probe angle of a flexible ureteroscope according to claim 4, characterized in that: The extracted axial torsion factor and local density factor are input as feature vectors into the trained convolutional neural network model. The convolutional neural network model automatically analyzes the combination relationship of the axial torsion factor and the local density factor, generates a topological complexity coefficient for characterizing structural complexity, and quantitatively evaluates the structural complexity of the target detection area based on the topological complexity coefficient.

6. A method for calibrating the probe angle of a flexible ureteroscope according to claim 5, characterized in that: After obtaining the topological complexity coefficient generated by the convolutional neural network model when evaluating the structural complexity of the target detection area, the topological complexity coefficient is compared and analyzed with the pre-set topological complexity coefficient reference threshold to divide the target detection area. The specific division steps are as follows: If the topological complexity coefficient is greater than the topological complexity coefficient reference threshold, the target detection area currently observed by the probe is divided into a high-complexity tortuous area; If the topology complexity coefficient is less than or equal to the topology complexity coefficient reference threshold, the target detection area currently observed by the probe is divided into a regular area.

7. A method for calibrating the probe angle of a flexible ureteroscope according to claim 6, characterized in that: For the target detection area divided into regular areas, the initial adjustment time remains unchanged, that is: , where: is the actual adjustment time for the regular area, is the preset initial adjustment time; For target detection areas classified as high-complexity tortuous areas, the adjustment time extension mechanism is triggered. The initial adjustment time is adaptively extended through the convolutional neural network evaluation results. The extended adjustment time is dynamically adjusted according to the relationship between the topological complexity coefficient and the topological complexity coefficient reference threshold. The specific formula is as follows: , where: is the actual adjustment time of the high-complexity tortuous area, after adaptive extension, is the topological complexity coefficient of the target detection area calculated by the convolutional neural network, which indicates the structural complexity of the target detection area. is the reference threshold of topological complexity, which is used as a standard value to adjust the extension ratio. k is the adjustment coefficient, which indicates the influence of the topological complexity coefficient on the extension of the adjustment time. .

8. A method for calibrating the probe angle of a flexible ureteroscope according to claim 4, characterized in that: For the target detection area, the specific steps of using feature engineering technology to process the degree of ureteral rotation on its axis and generate the axial torsion factor are as follows: Extract the local rotation angle of each segment from the ureter image of the target detection area; The rotation angle is calculated by tracking the change of the tangential direction of the ureter along the ureteral axis. To this end, two adjacent points are set and The vector between ,in: ; By calculating the angle change between adjacent vectors, the local rotation angle is obtained. The calculation expression of the local rotation angle is: , where: is the local rotation angle of the i-th segment, reflecting the degree of rotation of each segment, Represents a vector and vector The dot product of and Represents vectors and vector The model; According to the local rotation angle of each segment Calculate the axial torsion factor. The axial torsion factor measures the degree of axial torsion by the weighted cumulative value of the local rotation angle and considers the influence of local tortuosity on the total rotation complexity. The calculation expression is: , where: is the axial torsion factor of the target detection area, which represents the overall complexity of the rotation within the target detection area. are two adjacent points in the i-th segment and The axial distance between them reflects the length of each segment. is a positive adjustment parameter used to balance the relationship between the local rotation angle and the axial distance, and n is the total number of segments considered.

9. A method for calibrating the probe angle of a flexible ureteroscope according to claim 4, characterized in that: For the target detection area, the specific steps of using feature engineering technology to process the structural density of the ureter in the local area to generate the local density factor are as follows: Perform image enhancement and binarization on the target detection image obtained by the soft mirror imaging unit to obtain a binary structure image with clear structural boundaries , where a pixel value of 1 indicates the presence of ureteral tissue structure, and 0 indicates the absence of structure; Define a local sliding window with a radius of r centered at any reference point in the target detection area , and based on the local sliding window The distribution of internal structure pixels is used to calculate the structure occupancy rate of the local sliding window area. The calculation formula is as follows: , where: is the total number of pixels in the local sliding window; is the structural occupancy rate within the local sliding window area, which represents the degree of structural filling within the local sliding window range; Set the first scale radius With the second scale radius , respectively obtain the first scale radius With the second scale radius The structural occupancy of and , based on the first scale radius , the second scale radius , structural occupancy and , construct a multi-scale disturbance response model and generate a local density factor for quantifying the local tortuosity complexity. The generation expression of the local density factor is as follows: , where: is the local density factor, and are nonlinear amplification factors, respectively, used to enhance the response sensitivity to the absolute density and scale disturbance ratio of the local structure.

Images