A colonoscope pre-examination state evaluation system based on bowel sound monitoring

Abstract

The present application provides a colonoscopy intestinal pre-examination state evaluation system based on bowel sounds monitoring, and relates to the technical field of data processing, which comprises: an acquisition module for acquiring bowel sound signals; an enhancement module for processing the bowel sound signals to obtain enhanced bowel sound signals; an analysis module for performing frequency spectrum analysis on the enhanced bowel sound signals, extracting a plurality of main peak frequencies, and obtaining a peak frequency histogram through histogram statistics; a calculation module for calculating the similarity between the peak frequency histogram and a baseline window histogram, and judging whether the similarity is greater than a threshold; if yes, executing an extraction module; otherwise, executing a triggering module; the extraction module is used for extracting multi-dimensional dynamic spectrum features of the enhanced bowel sound signals; an output module is used for inputting the multi-dimensional dynamic spectrum features into a network model, outputting a pre-examination preparation result, and executing the triggering module; the triggering module is used for triggering a prompt mechanism and returning to the acquisition module when the preparation is not sufficient.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a pre-colonoscopy bowel condition assessment system based on bowel sound monitoring. Background Technology

[0002] Colonoscopy is an important diagnostic tool for digestive system diseases. Its imaging quality is closely related to the adequacy of the patient's bowel preparation before the examination. Bowel sounds, as an important physiological signal reflecting gastrointestinal motility, are strongly correlated with the efficiency of intestinal contents propulsion and the degree of intestinal cleanliness. Therefore, pre-examination status assessment based on bowel sound monitoring has gradually attracted attention.

[0003] Current methods for assessing bowel preparation before colonoscopy primarily rely on patient-reported bowel movements and the experience of healthcare professionals. The effectiveness of bowel preparation is estimated by observing external characteristics such as stool color, frequency, and degree of abdominal distension. While quantitative scales like the Boston Bowel Preparation Scale (BBPS) are commonly used clinically, these scores are typically only applied during the colonoscopy procedure and cannot provide dynamic preoperative monitoring. Recent studies have explored the use of wearable devices to collect bowel sound signals for time-domain or frequency-domain assessment of bowel motility.

[0004] However, existing studies based on bowel sounds often employ simple threshold or single-feature detection methods, which struggle to effectively distinguish between background noise, body movement interference, and genuine bowel sounds. This results in low signal recognition accuracy and an inability to reliably reflect the true state of bowel activity in patients. Furthermore, current technologies generally lack mechanisms to dynamically link changes in bowel sounds with the bowel preparation process, failing to automatically adjust the preparation procedure based on real-time monitoring results. This leaves some patients without timely intervention when bowel motility or cleansing is insufficient. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a colonoscopy bowel condition assessment system based on bowel sound monitoring. This system addresses the shortcomings of existing bowel sound-based studies, which often employ simple threshold or single-feature detection methods, making it difficult to effectively distinguish between background noise, body movement interference, and true bowel sounds. This results in low signal recognition accuracy and an inability to reliably reflect the patient's true bowel activity. Furthermore, existing technologies generally lack a mechanism to dynamically correlate bowel sound changes with the bowel preparation process, failing to automatically adjust the preparation procedure based on real-time monitoring results. This leads to the technical problem that some patients with insufficient bowel motility or inadequate bowel cleansing cannot receive timely intervention.

[0006] A first aspect of this invention provides a pre-colonoscopy bowel condition assessment system based on bowel sound monitoring, comprising: The acquisition module is used to acquire bowel sound signals from the pre-intestinal examination window; An enhancement module is used to extract effective segments from the bowel sound signal using an enhancement algorithm based on PNCC and artificial neural networks to obtain an enhanced bowel sound signal. The analysis module is used to perform spectral analysis on the enhanced bowel sound signal, extract multiple main peak frequencies of the enhanced bowel sound signal, and perform histogram statistics on each main peak frequency to obtain a peak frequency histogram characterizing the spectral distribution characteristics of the bowel sound signal. The calculation module is used to calculate the peak frequency similarity between the peak frequency histogram and the baseline window peak frequency histogram, and determine whether the peak frequency similarity is greater than the peak frequency similarity threshold; if so, the extraction module is executed; otherwise, the result of insufficient preparation before intestinal examination is returned, and the triggering module is executed. The extraction module is used to extract features from the enhanced bowel sound signal using the improved MFCC feature extraction method to obtain multidimensional dynamic spectral features; The output module is used to input the multidimensional dynamic spectral features into a pre-constructed dilated temporal convolutional network model, output the pre-colonoscopy intestinal examination preparation results, and execute the trigger module; The trigger module is used to trigger a prompting mechanism if the colonoscopy bowel preparation result is insufficient, suggesting that the user drink a small amount of warm water and wait for the bowel sounds to increase before returning to the acquisition module for re-testing.

[0007] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, by introducing a signal enhancement algorithm based on PNCC and artificial neural networks, the signal-to-noise ratio and extraction efficiency of bowel sound segments are effectively improved. Combining spectral peak extraction, histogram extraction, and improved MFCC extraction of multi-dimensional dynamic spectral features enhances the ability to express bowel sounds in the frequency domain, improving the system's resistance to background noise and body movement interference. Simultaneously, the spectral similarity-based evaluation mechanism enables real-time judgment of bowel preparation status, and the introduction of the dilated temporal convolutional network model improves the overall accuracy and stability of the evaluation. When the evaluation result indicates insufficient preparation, a prompting mechanism guides the user to adjust the status and automatically restarts the detection process, effectively reducing the risk of examination failure due to insufficient preparation and improving the intelligence and closed-loop control level of the entire bowel preparation process. Attached Figure Description

[0008] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0009] Figure 1 This is a schematic diagram of the structure of a colonoscopy intestinal condition assessment system based on bowel sound monitoring, provided in an embodiment of the present invention. Detailed Implementation

[0010] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0011] The pre-colonoscopy intestinal condition assessment system based on bowel sound monitoring, provided by the present invention, will be described in detail below with reference to the accompanying drawings and through specific embodiments and application scenarios.

[0012] Reference manual attached Figure 1 The diagram shows a structural schematic of a colonoscopy intestinal condition assessment system based on bowel sound monitoring provided by an embodiment of the present invention.

[0013] This invention provides a colonoscopy pre-examination bowel condition assessment system 10 based on bowel sound monitoring, which may include the following: The acquisition module 101 is used to acquire the bowel sound signal from the pre-intestinal examination window.

[0014] Specifically, a flexible electronic microphone patch is attached to the lower right abdomen of the subject (e.g., 9-11 cm from the navel). The patch is fixed to the skin surface using a medical-grade skin adhesive material, ensuring stable coupling of bowel sound signals to the microphone. The built-in microphone on the patch continuously or intermittently collects bowel sound signals at a sampling rate of 4 kHz. The collected digital signals are buffered locally and then transmitted to a mobile terminal or a higher-level analysis system via low-power wireless communication for subsequent signal enhancement, feature extraction, and bowel condition assessment analysis.

[0015] The enhancement module 102 is used to extract effective segments from the bowel sound signal using an enhancement algorithm based on PNCC and artificial neural networks to obtain an enhanced bowel sound signal.

[0016] Among them, PNCC (Power Normalized Cepstral Coefficients) is a robust auditory feature extraction method. By simulating the human ear's processing mechanism of speech signals in complex acoustic environments, it enhances the ability to suppress background noise and redundant interference, and can more effectively extract key spectral structures in non-speech short-time non-stationary signals such as bowel sounds.

[0017] Among them, artificial neural networks are nonlinear information processing models inspired by biological nervous systems. They achieve adaptive mapping and classification of complex input features through the connection and weight learning between multi-layer neuron nodes.

[0018] It should be noted that in this invention, ANN is used to determine the validity of bowel sound signal segments. Its input is the high-dimensional acoustic features extracted by PNCC, and its output is the classification probability of whether the signal segment belongs to a valid bowel sound. This model can be trained using the backpropagation algorithm, iteratively optimizing the weight parameters on existing bowel sound samples, thereby possessing good generalization ability and recognition accuracy, effectively improving the accuracy of subsequent bowel sound event extraction.

[0019] In one possible implementation, the enhancement module 102 is specifically used for: The bowel sound signal is segmented according to a fixed duration to obtain multiple bowel sound signal segments.

[0020] It should be noted that the fixed duration of 2.56 seconds is beneficial for subsequent statistics and feature learning.

[0021] Power-normalized acoustic cepstral features were extracted from each segment of the bowel sound signal.

[0022] In this embodiment of the invention, the bowel sound signal is segmented into segments of a fixed duration of 2.56 seconds, and the power-normalized acoustic cepstral feature (PNCC) of each segment is extracted. This helps to capture the key spectral features and energy distribution characteristics of bowel sounds while preserving the temporal context. This segmentation length is adapted to the average duration of bowel sounds and forms a consistent, dimensionally unified feature input sequence, facilitating subsequent effectiveness classification and judgment using an artificial neural network model, thereby improving the overall recognition accuracy and system robustness.

[0023] Each power-normalized acoustic cepstral feature is input into a pre-constructed artificial neural network model, which outputs the classification probability of each bowel sound signal segment as a valid bowel sound fragment.

[0024] Segments of bowel sound signals with a classification probability greater than the preset classification probability are selected to form a set of effective bowel sound segments.

[0025] It should be noted that those skilled in the art can set the preset classification probability according to actual needs, and this invention does not limit this.

[0026] Continuous or individually labeled effective segmented bowel sound fragments are combined into an independent bowel sound event to form an enhanced bowel sound signal.

[0027] In this embodiment of the invention, bowel sound signals are segmented according to fixed durations, and effective segments are identified using power-normalized acoustic cepstral features and an artificial neural network model. These segments are then merged into independent bowel sound events, significantly improving the targeting and accuracy of signal processing. On one hand, this process effectively filters out background noise and non-target interference, retaining the most diagnostically valuable real bowel sound segments. On the other hand, by unifying segment structure and event reconstruction, high-quality, standardized input data can be provided for subsequent spectral statistics, dynamic feature extraction, and intestinal state assessment models, thereby improving the robustness, real-time performance, and clinical discriminative ability of the overall system.

[0028] Analysis module 103 is used to perform spectral analysis on the enhanced bowel sound signal, extract multiple main peak frequencies of the enhanced bowel sound signal, and perform histogram statistics on each main peak frequency to obtain a peak frequency histogram characterizing the spectral distribution characteristics of the bowel sound signal.

[0029] Spectrum analysis refers to converting a time-domain signal (i.e., a signal that varies with time) into a frequency-domain signal to reveal its distribution characteristics at different frequency components. Through spectrum analysis, it is possible to identify which frequencies exist in the signal, the energy content of each frequency, and whether there are certain dominant frequencies or periodic structures.

[0030] In one possible implementation, the analysis module 103 is specifically used for: A third-order Butterworth bandpass filter was applied to filter the enhanced bowel sound signal.

[0031] The third-order Butterworth bandpass filter is a digital filter with a smooth frequency response and moderate transition bandwidth, often used for noise suppression and target frequency band extraction in signal processing. Its design feature is a maximally flat gain response within the passband, effectively preserving the main components of the signal without introducing fluctuation distortion.

[0032] It should be noted that in the processing of bowel sounds, using a third-order Butterworth bandpass filter to filter the signal in the 100-1000Hz range can significantly suppress low-frequency motion artifacts and high-frequency electrical noise, while preserving the typical frequency characteristics of bowel sounds, thus providing a clean and reliable input signal for subsequent spectrum analysis and feature extraction.

[0033] The filtered and enhanced bowel sound signal was divided into frames with a frame length of 64ms and a frame shift of 16ms, and a Hamming window of 128ms was applied to each frame.

[0034] The Hamming window is a commonly used windowing function for spectral analysis in signal framing, especially before performing the Fast Fourier Transform (FFT). Its function is to apply a gradually decreasing weight to both ends of each frame of signal to reduce spectral leakage.

[0035] It should be noted that, in order to transform the non-stationary bowel sound signal into a near-stationary short-time signal segment for easier spectral analysis and dominant frequency extraction, a 64ms frame length is used. This covers the basic period of a typical bowel sound signal, ensuring sufficient frequency resolution, while a 16ms frame shift provides high temporal resolution, helping to capture transient signal changes. Furthermore, a 128ms Hamming window is used to further expand the window function range and reduce spectral leakage caused by abrupt edge changes, thereby improving the spectral clarity and accuracy of dominant peak localization after the FFT transform. This processing step is one of the key pre-processes for bowel sound spectral feature extraction.

[0036] In this embodiment of the invention, the filtered enhanced bowel sound signal is framed with a frame length of 64ms and a frame shift of 16ms, and a Hamming window of 128ms is applied to each frame. This helps to divide the non-stationary bowel sound signal into multiple approximately stationary short frames, which facilitates frequency domain analysis.

[0037] Perform a Fast Fourier Transform on each windowed frame to extract the amplitude spectrum of each windowed frame.

[0038] The Fast Fourier Transform (FFT) is an efficient algorithm for implementing the Discrete Fourier Transform (DFT). It is used to convert time-domain signals into frequency-domain representations, thereby analyzing the presence and intensity of different frequency components in the signal. Compared to the high computational complexity required to directly compute the DFT, the FFT significantly reduces the computational load, making it particularly suitable for real-time spectral analysis of large-scale signals.

[0039] Based on each amplitude spectrum, the frequency component corresponding to the maximum amplitude is selected as the main peak frequency of the current frame.

[0040] Calculate the signal-to-noise ratio (SNR) of each main peak frequency in each bowel sound event, and select the main peak frequency corresponding to the maximum SNR as the main peak frequency of the corresponding bowel sound event.

[0041] Specifically, in each frame, the main peak frequency component corresponding to the maximum amplitude is found and its amplitude is recorded as "signal power"; then, a certain width of frequency band (such as ±50 Hz) is set around this frequency as "noise band", and the average amplitude of the frequency points other than the main peak in this band is calculated as "noise power"; finally, the ratio of the two (usually in logarithmic form, in dB) is used as the signal-to-noise ratio of the main peak frequency.

[0042] The peak frequencies of all bowel sounds within a preset time window are statistically analyzed using a histogram based on a preset bin width to obtain a peak frequency histogram.

[0043] The peak frequency histogram is a statistical graph used to describe the distribution characteristics of the dominant frequency in a signal, and is particularly suitable for analyzing the frequency characteristic changes of non-stationary, event-driven sound signals (such as bowel sounds). In this invention, it is used to quantify the probability of the occurrence of the dominant frequency of bowel sounds in different frequency bands, reflecting the spectral structure of intestinal activity.

[0044] Bin is a basic unit used to divide a continuous frequency range into several equally wide intervals.

[0045] It should be noted that those skilled in the art can set the size of the preset time window and the preset bin width according to actual needs, and this invention does not limit them.

[0046] In this embodiment of the invention, by performing a Fast Fourier Transform on each windowed signal to extract the amplitude spectrum, and further identifying the main peak frequency and calculating the signal-to-noise ratio, the most representative frequency components in each bowel sound event can be accurately extracted. Subsequently, all main peak frequencies are statistically analyzed using a histogram with a preset bin width to form a peak frequency histogram. This not only achieves structured modeling of the bowel sound spectrum distribution but also reflects the frequency domain characteristics of physiological states such as changes in intestinal contents and gas movement. This process facilitates quantitative comparison of bowel activity under different time windows, providing highly interpretable and stable frequency domain features for subsequent bowel clearance status assessment.

[0047] The calculation module 104 is used to calculate the peak frequency similarity between the peak frequency histogram and the baseline window peak frequency histogram, and to determine whether the peak frequency similarity is greater than the frequency similarity threshold. If so, the extraction module 105 is executed. Otherwise, the system returns "Insufficiently prepared results before intestinal examination" and the triggering module 107 is executed.

[0048] It should be noted that the peak frequency histogram of the baseline window represents the initial bowel sound frequency distribution before real-time bowel preparation.

[0049] In one possible implementation, the computing module 104 is specifically used for: The peak frequency histogram to be analyzed is paired with the peak frequency histogram of the baseline time window to determine the histogram vector of the analysis window and the histogram vector of the baseline window.

[0050] Optionally, the histogram vector of the analysis window is: ,in, This represents the histogram vector of the analysis window. This represents the frequency count of the i-th bin in the analysis window, where i = 1, 2, ..., N, and N represents the total number of bins in the histogram. The histogram vector for the baseline window is: ,in, Let represent the histogram vector of the baseline window, where represents , represents , and represents .

[0051] Calculate the mean and standard deviation of the histogram vector of the analysis window and the histogram vector of the baseline window, respectively.

[0052] The peak frequency similarity is calculated based on the mean and standard deviation of the histogram vectors of the analysis window and the baseline window.

[0053] in, This represents the peak frequency similarity, where N represents the total number of bins in the histogram. This represents the frequency count of the i-th bin in the analysis window. This represents the mean of the histogram vector of the analysis window. This represents the standard deviation of the histogram vector of the analysis window. This represents the frequency count of the i-th bin within the baseline window. This represents the mean of the baseline window histogram vector. This represents the standard deviation of the baseline window histogram vector.

[0054] Peak Frequency Histogram Similarity (PFHS) is an indicator used to measure the difference in the distribution of the main peak frequencies of bowel sounds within two time windows. It is mainly calculated based on the similarity between peak frequency histograms. This indicator can quantify whether the bowel sound spectrum changes significantly before and after bowel cleansing, thus helping to determine whether the intestines are adequately prepared.

[0055] Determine if the peak frequency similarity is greater than the peak frequency similarity threshold. If yes, execute extraction module 105. Otherwise, return that the intestinal examination results were not adequately prepared and execute trigger module 107.

[0056] It should be noted that those skilled in the art can set the frequency similarity threshold according to actual needs, and this invention does not impose any limitations on it.

[0057] In this embodiment of the invention, by calculating the peak frequency histogram similarity (PFHS) between the current analysis window and the baseline window and comparing it with a set frequency similarity threshold, a preliminary determination can be made as to whether there has been a significant change in the bowel cleansing state. This process does not rely on a single bowel sound event or frequency value, but rather assesses the degree of shift in the dominant frequency distribution pattern of bowel sounds during the bowel cleansing process from the perspective of the overall spectral structure, thus exhibiting stronger robustness and individual adaptability.

[0058] Furthermore, if PFHS is below the threshold, it indicates that the current bowel sound frequency pattern has significantly deviated from the initial state, suggesting that the bowel preparation effect may be good; conversely, if PFHS is still similar to the baseline, it indicates that the intestinal state has not changed sufficiently, which is judged as insufficient cleaning and triggers the feedback mechanism. As a screening checkpoint before subsequent dynamic feature analysis and deep model evaluation, this step has the advantages of being real-time, lightweight, and interpretable, and is a key initial screening step in the bowel preparation adequacy assessment process.

[0059] The extraction module 105 is used to extract features from the enhanced bowel sound signal using the improved MFCC feature extraction method to obtain multidimensional dynamic spectral features.

[0060] It should be noted that the improved MFCC feature extraction method introduces first-order difference (ΔMFCC) and second-order difference (ΔMFCC) based on the traditional Mel-frequency cepstral coefficients (MFCC). 2 This method uses MFCC (Multiple Mode and Cosine Curse) coefficients to more comprehensively characterize the dynamic changes of signals over time. First, it constructs an energy spectrum map through pre-emphasis, framing, windowing, Fast Fourier Transform, and Mel filter banks. Then, it extracts basic MFCC features through logarithmic compression and Discrete Cosine Transform. Subsequently, it performs first-order and second-order difference calculations on the MFCCs to capture the trend and rate of change of the features over time. Finally, it concatenates the three types of features into a high-dimensional dynamic spectral feature vector. This method can simultaneously preserve the static spectral structure and dynamic change patterns of bowel sounds, improving the model's ability to distinguish non-stationary signals in time series and providing richer and more robust input representations for deep models such as convolutional recurrent networks.

[0061] In one possible implementation, the extraction module 105 is specifically used for: A first-order FIR pre-emphasis filter is applied to the enhanced bowel sound signal for pre-emphasis processing:

[0062] in, This represents the value of the enhanced bowel sound signal after pre-emphasis processing at the nth sampling point. This represents the value of the original enhanced bowel sound signal at the nth sampling point. Indicates the pre-emphasis coefficient. This represents the value of the original enhanced bowel sound signal at the (n-1)th sampling point.

[0063] The first-order FIR pre-emphasis filter is a simple and effective high-pass filtering operation, often used in the pre-processing stage of audio signal processing. Its purpose is to emphasize high-frequency components and suppress low-frequency energy shift, thereby improving the effect of subsequent feature extraction.

[0064] Using a frame length of 64ms and a frame shift of 16ms, the enhanced bowel sound signal after pre-emphasis processing was divided into frames to obtain multiple time-domain bowel sound signal segments after framing.

[0065] For each frame of time-domain bowel sound signal segment, a Fast Fourier Transform is performed to extract multiple power spectra. These power spectra are then weighted and summed using a set of triangular filters with equal Mel frequency intervals to obtain the energy characteristics.

[0066] in, This represents the energy characteristic of the m-th Mel-triangle filter output, where K represents the total number of frequency points. This represents the power value at the k-th frequency point in the power spectrum. This represents the weighting coefficient of the m-th Mel triangular filter at the k-th frequency point.

[0067] Among them, the triangular filter with equal Mel frequency intervals is a filter bank designed to simulate the human ear's auditory perception mechanism and is widely used in the process of MFCC (Mel frequency cepstral coefficients) feature extraction. After mapping the frequency axis to the Mel scale, this filter bank divides the frequency bands in a perceptually "equal distance" manner, and constructs a centrally symmetric triangular filter in each frequency band for weighting spectral energy.

[0068] It should be noted that each triangular filter only responds to the frequency components of its covered band, and the response value gradually decreases as the frequency moves away from the center frequency. This design not only preserves the physiological characteristics of the human ear, which is more sensitive to low frequencies and less sensitive to high frequencies, but also extracts spectral energy distribution features with stable speech / sound structures while compressing dimensions. This is one of the key steps in generating MFCCs and also demonstrates good discriminative ability in processing non-stationary sound signals such as bowel sounds.

[0069] Logarithmic compression of the energy features yields the logarithmic energy sequence:

[0070] in, Represents a logarithmic function.

[0071] Performing a discrete cosine transform on the logarithmic energy sequence yields the baseline MFCC eigenvectors:

[0072] in, Let M represent the nth-dimensional baseline MFCC eigenvector, where M represents the number of Mel filters. Represents the cosine function. Let π represent pi, n represent the dimension index of the current discrete cosine transform, and L represent the final number of MFCC dimensions retained.

[0073] The Discrete Cosine Transform (DCT) is a mathematical transformation commonly used in signal processing and feature compression. It transforms the energy distribution in the time or frequency domain into a set of coefficients with good concentration characteristics, making it particularly suitable for preserving low-frequency information while suppressing high-frequency noise.

[0074] Calculate the first-order and second-order difference eigenvectors of the baseline MFCC eigenvectors:

[0075]

[0076] in, This represents the baseline MFCC feature vector of frame t. Let represent the first-order difference vector of the baseline MFCC eigenvector of frame t, where i represents the index step size within the difference calculation window, k represents the difference window length, and t represents the index of the current frame. This represents the MFCC feature vector extracted in frame t+i. This represents the MFCC feature vector extracted in frame ti. Let represent the second-order difference vector of the baseline MFCC eigenvector of frame t.

[0077] It should be noted that the intermediate difference window length k = 2 or 3, and the specific value can be adjusted.

[0078] By concatenating the baseline MFCC feature vector, the first-order difference feature vector, and the second-order difference feature vector, a multidimensional dynamic spectral feature is obtained:

[0079] in, This represents the multidimensional dynamic spectral characteristics of the t-th frame.

[0080] Output module 106 is used to input multidimensional dynamic spectral features into a pre-built dilated temporal convolutional network model, output the pre-colonoscopy intestinal examination preparation results, and execute trigger module 107.

[0081] In one possible implementation, the output module 106 is specifically used for: Organize the multidimensional dynamic spectral feature vectors into a two-dimensional input feature sequence in chronological order:

[0082] in, Let X represent the initial input to the dilated temporal convolutional network, and let X represent the two-dimensional input feature sequence. Let d represent the multidimensional dynamic spectral features of frame t, where t = 1, 2, ..., T, T represents the total number of frames within the current analysis time window, and d represents the feature dimension of each frame.

[0083] The two-dimensional input feature sequence is used as the initial input of the dilated temporal convolutional network model, and multi-layer dilated convolution and residual connection recursively computed for each frame sequence.

[0084]

[0085]

[0086] in, This indicates that the activation of layer l in the j-th TCN block is... The output at time t represents time t, j represents the j-th TCN block, l represents the l-th layer in the current block, and f represents the non-linear activation function. This represents the first weight matrix of the convolutional layer. Indicates the j-th TCN block. Layer in The input feature vector at time t, where s represents the expansion step size. This represents the second weight matrix of the convolutional layer. Indicates the j-th convolutional block. The input feature vector of the layer at time t, where b represents the dilated convolution bias term. V represents the final output feature of layer l in the j-th TCN block, and V represents the projection matrix of the residual connection. Let represent the input features of layer l in the j-th TCN block at time t. This represents the residual connection bias term.

[0087] The output features of each frame from the last layer are fused using skip connections, and a higher-order state representation is obtained through nonlinear mapping, resulting in multiple classification and discriminative fused features:

[0088]

[0089] in, This represents the frame-level features after all blocks have been fused. Let B represent the activation function, and let B represent the number of TCN blocks. This represents the output feature of the last layer in the j-th TCN block at time t. Classification discriminant features representing higher-order semantic expressions This represents the fully connected weight matrix of the mapping layer. This represents the mapping layer bias term.

[0090] Based on the fusion features of various classification criteria, the predicted probability of the intestinal preparation state category is output through a softmax layer:

[0091] in, This represents the predicted probability of the gut preparation state category. This represents the activation function. Represents the classifier weight matrix. This represents the classifier bias term.

[0092] The average maximum value strategy is used to merge the various prediction probabilities to obtain the final prediction probability.

[0093] The colonoscopy intestinal tract preparation results are determined based on the final predicted probability.

[0094] In one possible implementation, the pre-colonoscopy preparation results include adequate preparation and inadequate preparation. The output module 106 is further specifically used for: Determine if the final predicted probability is greater than the probability threshold. If yes, the colonoscopy preparation is deemed adequate. Otherwise, the colonoscopy preparation is deemed inadequate.

[0095] It should be noted that those skilled in the art can set the preset probability threshold according to actual needs, and this invention does not limit this.

[0096] In this embodiment of the invention, by inputting a multidimensional dynamic spectral feature sequence into a dilated temporal convolutional network and sequentially performing multi-layer dilated convolution, residual connections, higher-order semantic fusion, and softmax prediction, the long-term and short-term dependencies and local dynamic patterns of bowel sound signals in the time dimension can be fully extracted, thereby achieving accurate classification and discrimination of bowel preparation status. This process not only possesses deep feature representation capabilities but also utilizes dilated convolution to expand the receptive field and avoid information sparsity. Simultaneously, it enhances the model's stability and generalization performance through skip connections and probabilistic fusion. The final output prediction result can reflect the degree of bowel cleanliness under the current window in real time and stably, which is one of the core links in achieving intelligent, end-to-end pre-detection assessment in this invention.

[0097] Trigger module 107 is used to trigger a prompting mechanism when the result of pre-colonoscopy bowel preparation is insufficient, suggesting that the user drink a small amount of warm water and wait for the bowel sounds to increase before returning to the acquisition module 101 for re-testing.

[0098] It should be noted that a small amount of warm water is specifically 200ml.

[0099] In one possible implementation, the trigger module 107 determines whether bowel sounds are enhanced in the following way: Reacquire bowel sound signals and compare them with those from the insufficient bowel sound state. If the number of effective bowel sound segments and the peak frequency energy increase, the bowel sounds are considered enhanced, and the next round of evaluation is executed. Otherwise, maintain the status quo and wait for the preset time before reacquiring bowel sound signals.

[0100] It should be noted that those skilled in the art can set the preset duration according to actual needs, and this invention does not impose any limitations on it.

[0101] In this embodiment of the invention, a warm water stimulation and reassessment mechanism is introduced when bowel preparation for colonoscopy is insufficient. The decision to proceed with the next assessment round is made by comparing changes in bowel sound signals before and after the procedure, demonstrating significant practical value and clinical applicability. On one hand, this method stimulates intestinal peristalsis through non-pharmacological intervention, making it both safe and convenient for users. On the other hand, based on the dynamic comparison of the number of effective bowel sound segments and the energy of the main peak frequency, real-time monitoring and quantitative determination of bowel activity levels are achieved, helping to avoid ineffective waiting or unnecessary examination delays. Furthermore, the ability to set flexible waiting times enhances the system's adaptability to individual differences, making the assessment process more intelligent, flexible, and humane. This strategy can form a closed-loop feedback control, providing a reliable dynamic adjustment mechanism for the pre-colonoscopy bowel state.

Claims

1. A pre-colonoscopy bowel condition assessment system based on bowel sound monitoring, characterized in that, include: The acquisition module is used to acquire bowel sound signals from the pre-intestinal examination window; An enhancement module is used to extract effective segments from the bowel sound signal using an enhancement algorithm based on PNCC and artificial neural networks to obtain an enhanced bowel sound signal. The analysis module is used to perform spectral analysis on the enhanced bowel sound signal, extract multiple main peak frequencies of the enhanced bowel sound signal, and perform histogram statistics on each main peak frequency to obtain a peak frequency histogram characterizing the spectral distribution characteristics of the bowel sound signal. The calculation module is used to calculate the peak frequency similarity between the peak frequency histogram and the baseline window peak frequency histogram, and to determine whether the peak frequency similarity is greater than the peak frequency similarity threshold. If so, execute the extraction module; Otherwise, if insufficient results were prepared before the intestinal examination, the trigger module will be executed; The extraction module is used to extract features from the enhanced bowel sound signal using the improved MFCC feature extraction method to obtain multidimensional dynamic spectral features; The output module is used to input the multidimensional dynamic spectral features into a pre-constructed dilated temporal convolutional network model, output the pre-colonoscopy intestinal examination preparation results, and execute the trigger module; The trigger module is used to trigger a prompting mechanism if the colonoscopy bowel preparation result is insufficient, suggesting that the user drink a small amount of warm water and wait for the bowel sounds to increase before returning to the acquisition module for re-testing.

2. The colonoscopy bowel condition assessment system based on bowel sound monitoring according to claim 1, characterized in that, The enhancement module is specifically used for: The bowel sound signal is segmented according to a fixed duration to obtain multiple bowel sound signal segments; Power-normalized acoustic cepstral features were extracted from each of the bowel sound signal segments. Each of the power-normalized acoustic cepstral features is input into a pre-constructed artificial neural network model, and the classification probability of each of the bowel sound signals being segmented into a valid bowel sound segment is output. Segments of bowel sound signals with a classification probability greater than a preset classification probability are selected to form a set of effective bowel sound segments; Continuous or individually labeled valid segmented bowel sound fragments are combined into an independent bowel sound event to form the enhanced bowel sound signal.

3. The colonoscopy bowel condition assessment system based on bowel sound monitoring according to claim 1, characterized in that, The analysis module is specifically used for: The enhanced bowel sound signal was filtered using a third-order Butterworth bandpass filter. The filtered and enhanced bowel sound signal was divided into frames with a frame length of 64ms and a frame shift of 16ms, and a Hamming window of 128ms was applied to each frame. Perform a fast Fourier transform on each windowed frame to extract the amplitude spectrum of each windowed frame; Based on each amplitude spectrum, the frequency component corresponding to the maximum amplitude is selected as the main peak frequency of the current frame; Calculate the signal-to-noise ratio of each of the main peak frequencies in each bowel sound event, and select the main peak frequency corresponding to the maximum signal-to-noise ratio as the main peak frequency of the corresponding bowel sound event. The peak frequency histogram is obtained by statistically analyzing the main peak frequencies of all bowel sound events within a preset time window according to a preset bin width.

4. The colonoscopy bowel condition assessment system based on bowel sound monitoring according to claim 1, characterized in that, The calculation module is specifically used for: The peak frequency histogram to be analyzed is paired with the peak frequency histogram of the baseline time window to determine the histogram vector of the analysis window and the histogram vector of the baseline window. Calculate the mean and standard deviation of the histogram vector of the analysis window and the histogram vector of the baseline window, respectively; The peak frequency similarity is calculated based on the mean and standard deviation of the histogram vector of the analysis window and the mean and standard deviation of the histogram vector of the baseline window. Determine whether the peak frequency similarity is greater than the peak frequency similarity threshold; If so, execute the extraction module; Otherwise, if insufficient results are prepared before the intestinal examination, the trigger module will be executed.

5. The colonoscopy bowel condition assessment system based on bowel sound monitoring according to claim 4, characterized in that, The peak frequency similarity is calculated as follows: ； in, This represents the peak frequency similarity, where N represents the total number of bins in the histogram. This represents the frequency count of the i-th bin in the analysis window. This represents the mean of the histogram vector of the analysis window. This represents the standard deviation of the histogram vector of the analysis window. This represents the frequency count of the i-th bin within the baseline window. This represents the mean of the baseline window histogram vector. This represents the standard deviation of the baseline window histogram vector.

6. The colonoscopy bowel condition assessment system based on bowel sound monitoring according to claim 1, characterized in that, The extraction module is specifically used for: The enhanced bowel sound signal is pre-emphasized by applying a first-order FIR pre-emphasis filter; Using a frame length of 64ms and a frame shift of 16ms, the enhanced bowel sound signal after pre-emphasis processing was divided into frames to obtain multiple time-domain bowel sound signal segments after framing. For each frame of the time-domain bowel sound signal segment, perform a fast Fourier transform to extract multiple power spectra, and then use a set of triangular filters with equal Mel frequency intervals to perform a weighted summation of each power spectrum to obtain energy characteristics; The energy features are subjected to logarithmic compression to obtain a logarithmic energy sequence; The discrete cosine transform is performed on the logarithmic energy sequence to obtain the baseline MFCC eigenvector; Calculate the first-order difference eigenvector and the second-order difference eigenvector of the reference MFCC eigenvector; The reference MFCC feature vector, the first-order difference feature vector, and the second-order difference feature vector are concatenated to obtain the multidimensional dynamic spectrum feature.

7. The colonoscopy bowel condition assessment system based on bowel sound monitoring according to claim 6, characterized in that, The multidimensional dynamic spectral characteristics: ； in, This represents the multidimensional dynamic spectral characteristics of the t-th frame. This represents the baseline MFCC feature vector of frame t. Let represent the first-order difference vector of the baseline MFCC eigenvectors in frame t. Let represent the second-order difference vector of the baseline MFCC eigenvector of frame t.

8. The colonoscopy bowel condition assessment system based on bowel sound monitoring according to claim 1, characterized in that, The output module is specifically used for: The multidimensional dynamic spectral feature vectors are organized into a two-dimensional input feature sequence in chronological order. The two-dimensional input feature sequence is used as the initial input of the dilated temporal convolutional network model, and multi-layer dilated convolution and residual connection recursively calculated for each frame sequence; The output features of each frame of the last layer are fused by skip connections, and a higher-order state representation is obtained through nonlinear mapping, resulting in multiple classification and discrimination fusion features. Based on the various classification and discrimination fusion features, the predicted probability of the intestinal preparation state category is output through the Softmax layer; The average maximum value strategy is used to fuse the various predicted probabilities to obtain the final predicted probability; Based on the final predicted probability, the results of pre-colonoscopy bowel preparation are determined.

9. The colonoscopy bowel condition assessment system based on bowel sound monitoring according to claim 8, characterized in that, The results of pre-colonoscopy bowel preparation include adequate preparation and inadequate preparation; the output module is further specifically used for: Determine whether the final predicted probability is greater than the probability threshold; if so, determine that the preparation for colonoscopy is sufficient; otherwise, determine that the preparation for colonoscopy is insufficient.

10. The colonoscopy bowel condition assessment system based on bowel sound monitoring according to claim 1, characterized in that, The specific method by which the trigger module determines whether bowel sounds are enhanced is as follows: The bowel sound signal is reacquired and compared with the bowel sound signal in the insufficient state. If the number of effective bowel sound segments and the main peak frequency energy increase, the bowel sound is determined to be enhanced, and the next round of evaluation process is executed. Otherwise, the bowel sound signal is reacquired after waiting for a preset time.

Images