Intelligent monitoring and early warning system for dynamic change of cerebrospinal fluid leakage

Abstract

The invention relates to the technical field of medical image processing and intelligent medical monitoring, in particular to an intelligent monitoring and early warning system for dynamic changes of cerebrospinal fluid leakage. The system comprises a time sequence image management module which takes a leakage point image coordinate P0 in reference input as a space anchor point, performs non-rigid registration on a multi-time-point follow-up visit image and a reference image, and outputs a time sequence image sequence in space-time alignment; the dynamic change quantization module is used for generating a multi-dimensional quantization parameter set through fusion calculation of morphological, density and texture features based on the sequence and the reference segmentation mask M0; and the intelligent early warning judgment module is used for matching the parameter set with a preset early warning rule base containing multi-parameter coupling logic and outputting a graded early warning signal. The cerebrospinal fluid leakage monitoring method based on the multi-parameter fusion quantification realizes the technical span of the cerebrospinal fluid leakage from static analysis to dynamic monitoring, and improves the accuracy, objectivity and efficiency of state perception through multi-parameter fusion quantification and intelligent judgment.

Description

Technical Field

[0001] This invention relates to the fields of medical image processing and intelligent medical monitoring technology, and more specifically to an intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage. Background Technology

[0002] In existing technologies, there are methods and systems capable of intelligent identification and initial localization of cerebrospinal fluid leaks. For example, by combining multimodal image registration (such as the ANTs algorithm) with artificial intelligence models (such as nnUnet and YOLOv8), accurate segmentation of cerebrospinal fluid leak areas and localization of leak coordinates in single-time-point images have been achieved, effectively solving the problem of initial state identification and localization.

[0003] However, the imaging manifestations of cerebrospinal fluid leakage are a dynamic process. The aforementioned existing technologies and other similar solutions all focus on static, one-off analysis, failing to meet the technical requirements for long-term, dynamic, and quantitative tracking of the target area. Specifically, existing technologies have the following technical shortcomings: Lack of time series analysis capability: Existing solutions perform isolated analysis of single acquired images. Their system architecture cannot automatically correlate and compare sequential image data of the same target at multiple time points, thus failing to generate continuous data describing the evolution of the target.

[0004] Change assessment relies on manual comparison: For images from different time points, operators must manually compare and analyze them to qualitatively determine changes in the target area. This process is inefficient and highly susceptible to subjective interference, resulting in poor consistency and low reliability of the analysis results.

[0005] Lack of objective quantitative early warning mechanism: Because the system cannot automatically extract and calculate the quantitative evolution parameters of the target area (such as volume change rate and changes in image feature values), existing technologies cannot establish a data-based automated state assessment and alert mechanism. It is difficult to identify significant changes in the state in real time from a technical perspective.

[0006] The status assessment lacks continuous data support: the system cannot provide an objective, quantitative data chain based on time-series imagery for the overall status of the target area. Operators struggle to accurately grasp its dynamic details, making it impossible to provide efficient and reliable data support for subsequent comprehensive analysis. Summary of the Invention

[0007] In view of this, the present invention provides an intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage. The system aims to use the image coordinates and segmented regions of the leak point determined by the initial intelligent identification and localization as the benchmark input to construct an analytical model for quantitatively tracking its dynamic changes and executing intelligent early warning, thereby ultimately realizing intelligent monitoring and early warning of dynamic changes in cerebrospinal fluid leakage.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage, comprising:

[0009] In a specific feasible implementation,

[0010] In a specific feasible implementation,

[0011] Secondly, the present invention provides an intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage, applied to the aforementioned... Thirdly, this invention provides an intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage. Compared with existing technologies, the intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage described in this invention breaks through the existing technical paradigm of isolated analysis of images at a single time point. It innovatively constructs a dynamic monitoring and early warning system for a target area, which has the following beneficial effects: 1. A leap in system capabilities from "static analysis" to "dynamic monitoring" Existing technologies provide a "static snapshot" of a target at a "certain moment," while this invention generates a "dynamic evolution map" of the target in the "time dimension" by systematically introducing temporal image analysis. This transformation enables the system to upgrade from providing "spatial state" information to providing "spatiotemporal evolution" information, providing unprecedented continuous data support for a comprehensive understanding of the dynamic process of the target object, and achieving a fundamental leap in analytical capabilities from static to dynamic.

[0012] 2. High-precision variation quantization model based on "benchmark anchor point" and multi-parameter fusion This invention constructs a novel core technology model. Its innovation lies first in defining the initial "reference input" (i.e., coordinates P0 and segmentation mask M0) as a stable and unchanging spatial reference anchor point throughout the entire temporal analysis. This serves as the core driving force for high-precision temporal registration, solving the fundamental technical challenge of pixel-level accurate comparison of multi-source heterogeneous image data due to differences in acquisition parameters. Based on this, the invention creates a multi-dimensional quantitative analysis model that integrates morphology (volume change rate VCR), density (average grayscale change ΔMean), and structural complexity (texture entropy change ΔEntropy). This multi-feature fusion analysis method can more sensitively and comprehensively perceive and quantify subtle evolutions of the target, significantly improving the system's accuracy, robustness, and information richness in identifying dynamic changes.

[0013] 3. Achieve an automated and intelligent closed loop from "data" to "insight". This invention transforms advanced algorithm models into highly practical automated solutions through system-level integration. By integrating an "intelligent early warning and judgment module," the system automatically matches the aforementioned quantitative parameters with a preset, configurable rule base, enabling the automatic extraction of key changing trends from massive image data and the generation of tiered early warning signals. This design replaces the traditional model that relies on manual comparison and subjective judgment by operators, providing an objective, real-time, and consistent insight into state changes. Finally, the system integrates and outputs multi-dimensional data, images, and analysis conclusions through a "results visualization and reporting module," forming an end-to-end automated closed loop from data access, intelligent analysis, change perception to result generation, greatly improving information processing efficiency and the reliability of system output. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] Figure 1 This is a flowchart illustrating the overall workflow of the intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage as described in this invention.

[0016] Figure 2 This is a hardware architecture diagram of an intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage as described in this invention.

[0017] Figure 3 This is a schematic diagram of the registration process for the time-series image management module.

[0018] Figure 4 A flowchart for parameter calculation in the dynamic quantization module.

[0019] Figure 5 This is a logic diagram for rule matching in the intelligent early warning and judgment module. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention discloses an intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage, applied to the clinical monitoring and management of cerebrospinal fluid leakage. The system uses a set of structured "baseline inputs" as the foundation of the entire dynamic monitoring process. Through systematic analysis of multi-time-point sequence images, it achieves quantitative tracking and intelligent early warning of dynamic changes in the target area, including: System hardware architecture and data foundation, such as Figure 2 As shown.

[0022] The system described in this invention is deployed on a computing device that includes at least one processor, memory, and input / output interfaces. The system establishes a data connection with a hospital's Picture Archiving and Communication System (PACS) via the input / output interfaces to acquire and process medical image data.

[0023] This system relies on two types of core input data to function: 1. Baseline Input Data: This data originates from computed tomography (CT) or magnetic resonance imaging (MRI) images of the brain or spinal region acquired during the patient's initial examination. The following two structured data points are obtained through initial analysis: Leakage point spatial coordinates: a three-dimensional coordinate point in the patient coordinate system defined by the DICOM standard. Its numerical unit is millimeters (mm), and it is used to accurately identify the anatomical location of cerebrospinal fluid leaks.

[0024] Missing region segmentation mask: a binary image matrix that perfectly matches the spatial dimensions (length, width, number of layers) of the reference image. In this matrix, the set of pixels with a value of 1 represents the part identified by the algorithm as a "cerebrospinal fluid leakage area", while the pixels with a value of 0 represent the background or other tissue.

[0025] For example, a target detection model based on the YOLOv8 architecture can be used to initially locate the leak point to obtain the coordinates P0, and a segmentation model based on the nnU-Net architecture can be used to accurately segment the leak region to generate a mask M0.

[0026] Temporal imaging data: CT or MRI image sequences $I_1, I_2, ..., I_n$ acquired again at different follow-up time points $T_1, T_2, ..., T_n$ for the same anatomical sites of the same patient after the initial examination.

[0027] System functional modules and workflow like Figure 1 As shown, this system is logically composed of four sequentially cooperating functional modules.

[0028] 1. Time-series image management module like Figure 3 As shown, this module is responsible for receiving and preprocessing all input data, providing a precisely aligned image basis for subsequent quantitative analysis.

[0029] Data Interface Unit: Establishes a data connection with the hospital's Picture Archiving and Communication System (PACS) to automatically acquire the following data for a specified patient: Baseline input data includes baseline image $I_0$, cerebrospinal fluid leak image coordinates $P_0$, and leak region segmentation mask $M_0$.

[0030] Time-series image data: Images $I_t$ collected at follow-up time point $t$.

[0031] Temporal registration unit: This unit is crucial for ensuring the comparability of images at different time points. It performs the following operations: A three-dimensional region of interest (VOI) is defined with the reference leak point coordinates $P_0$ as the center.

[0032] The Symmetric Normalization (SyN) algorithm from the Advanced Normalization Tools (ANTs) package is invoked to compute the nonlinear spatial transformation field $T_{0 \rightarrowt}$ that aligns the spatial geometry of the follow-up image $I_t$ to the reference image $I_0$.

[0033] Using this transform field, the follow-up image $I_t$ is resampled to the reference image space to obtain a registered image $I_t'$ that is precisely aligned on the anatomical structure, i.e., $I_t' = T_{0 \rightarrow t} \circ I_t$.

[0034] 2. Dynamic Change Quantization Module like Figure 4 As shown, this module automatically calculates a series of objective quantitative indicators that characterize the dynamic changes of the leak area based on the registered image.

[0035] Volume change rate calculation unit: On the registered image $I_t'$, with $P_0$ as the center, the segmentation model (preferably using the nnU-Net framework) automatically generates the missing region segmentation mask $M_t$ for the current time point.

[0036] Calculate the physical volumes $V_0$ and $V_t$ represented by the masks $M_0$ and $M_t$. The volume calculation formula is: $V = N \times \Delta x \times \Delta y \times \Delta z$, where $N$ is the total number of pixels with a value of 1 in the mask, and $\Delta x, \Delta y, \Delta z$ are the pixel spacing and layer thickness of the image.

[0037] Calculate the volume change rate $VCR$: $VCR = \frac{V_t - V_0}{V_0} \times 100%$.

[0038] Image feature analysis unit: Within the common region defined by $M_0$ and $M_t$, calculate the changes in image grayscale features.

[0039] Calculate the average grayscale value change, denoted as Delta Mean. $\Delta Mean = \frac{1}{N_{union}} \sum_{p \in (M_0 \cup M_t)} I_t'(p) - \frac{1}{N_0} \sum_{q \in M_0} I_0(q)$ Where $N_{union}$ and $N_0$ are the total number of pixels in the union region and the reference mask region, respectively.

[0040] The texture feature changes are calculated by calculating the entropy change $\Delta Entropy$ of the gray-level co-occurrence matrix within the region. The entropy calculation formula is: $Entropy = -\sum_{i=1}^{N_g} \sum_{j=1}^{N_g} p(i,j)\log_2 p(i,j)$, where $p(i,j)$ is the normalized co-occurrence matrix value and $N_g$ is the number of gray levels.

[0041] 3. Intelligent Early Warning and Judgment Module like Figure 5 As shown, this module automatically judges the quantification results based on preset logical rules and generates early warning signals.

[0042] The module pre-stores a rule base, with rules in the form of "IF condition THEN conclusion".

[0043] Example rule: If VCR > 25%, then the warning level is "Significant Progress". IF (VCR>15%) AND (ΔEntropy>0.1) THEN Warning Level = "Progress Monitoring" If VCR < -20%, then the warning level is "Significant Absorption". This module receives parameters such as $VCR$, $Delta Mean$, and $DeltaEntropy$ from the dynamic change quantification module, matches them with the rule base, and outputs the corresponding warning level and description.

[0044] Furthermore, the aforementioned early warning rule base is not static but possesses dynamic self-learning and update capabilities. The system configuration is as follows: Receive feedback: Continuously receive confirmation or correction feedback from users (such as doctors) regarding the system's warning results, and use the feedback results as reward signals for reinforcement learning.

[0045] Model Adjustment: Based on accumulated feedback data, the system periodically optimizes and adjusts thresholds (such as 25% of VCR, 0.1 of Delta Entropy) or logical combinations in the rule base using reinforcement learning algorithms (e.g., Q-learning). Specifically, the system compares user feedback with the warning results triggered by the current rules to adjust rule parameters, making future warning outputs more consistent with clinical judgment.

[0046] Results: This allows the system's early warning judgments to continuously adapt to actual clinical diagnostic standards and the operating habits of different hospitals, achieving continuous improvement in personalization and precision.

[0047] 4. Results Visualization and Reporting Module This module integrates all data, analysis results, and early warning information to generate output that doctors can use directly.

[0048] Generate a dynamic trend chart, plotting curves for parameters such as $VCR$ with time as the horizontal axis.

[0049] Provides a multi-time point image comparison view, displays $I_0$ and $I_t'$ side by side, and overlays the outlines of $M_0$ and $M_t$ with different colors.

[0050] The system automatically generates structured monitoring reports, outputting them in PDF or HTML format. The reports include patient information, quantitative data at various time points, early warning conclusions, and key image screenshots.

[0051] 5. Data Quality Assessment Module This module is crucial for ensuring system reliability. Before the dynamic change quantization module begins its calculations, this module automatically evaluates the image quality of the registered image $I_t'$ within the neighborhood of the reference coordinate P0.

[0052] Evaluation metrics: The main evaluation metrics are signal-to-noise ratio (SNR) and local contrast. A low SNR may lead to feature extraction errors, while insufficient local contrast will affect the clarity of segmentation boundaries.

[0053] Evaluation process: This module calculates the ratio of the standard deviation of the signal intensity in the neighborhood to the standard deviation of the background area as the signal-to-noise ratio estimate; at the same time, it calculates the grayscale difference between the target area and the surrounding tissue area as the local contrast.

[0054] Decision and Feedback: If the calculated signal-to-noise ratio (SNR) (e.g., SNR below 20 dB) or local contrast (e.g., grayscale difference between target and background less than 100 HU) is lower than a preset quality threshold, this module will not pass the data to the subsequent quantization module. Instead, it will issue an instruction to the system to trigger a data re-acquisition process or send a "Poor image quality, review recommended" message to the operator, thereby ensuring the accuracy of the analysis results from the source. The quality threshold can be configured according to different image modalities (e.g., CT or MRI) and equipment performance.

[0055] The various embodiments described in this specification are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage, characterized in that, include: The temporal image management module is configured to use the coordinates P0 of the leaked image in the reference input as the spatial anchor point, and to perform non-rigid registration between the follow-up images acquired at different time points and the reference image, and output a temporally and spatially aligned temporal image sequence. The dynamic change quantization module is configured to generate a multi-dimensional quantization parameter set to characterize dynamic changes by fusing morphological, density, and texture features based on the time-series image sequence and the missing region segmentation mask M0 in the reference input. The intelligent early warning decision module is configured to match the multi-dimensional quantization parameter set with a preset early warning rule base, wherein the early warning rule base contains rules defined based on the logical combination of multiple parameters in the multi-dimensional quantization parameter set, and outputs corresponding graded early warning signals.

2. The intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage according to claim 1, characterized in that, The specific method by which the temporal image management module performs registration is as follows: Centered on the image coordinates P0 of the leak point, a three-dimensional region of interest is defined to drive registration optimization; Within the three-dimensional region of interest, a symmetric normalization transformation based on maximizing mutual information is used to solve the nonlinear spatial transformation field $T_{0 \rightarrow t}$ that maps the follow-up image It to the reference image I0 space.

3. The intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage according to claim 2, characterized in that, The time-series image management module is also configured to: After acquiring the follow-up image It, its initial similarity with the reference image I0 within the three-dimensional region of interest is automatically verified. If the initial similarity is lower than a preset threshold, the multi-resolution search strategy of the registration algorithm is adjusted first, and then the nonlinear spatial transformation field is solved.

4. The intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage according to claim 1, characterized in that, The multidimensional quantization parameter set includes at least the volume change rate (VCR), the average grayscale value change (\Delta Mean), and the texture entropy value change (\Delta Entropy).

5. The intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage according to claim 4, characterized in that, The volume change rate (VCR) is calculated in the following way: On the registered image $I_t'$, with the coordinate P0 as the search center, the segmentation model based on the nnU-Net architecture is called to generate the missing region segmentation mask Mt for the current time point; The volume change rate is calculated using the formula $VCR = \frac{V_t - V_0}{V_0} \times 100%$, where $V_0$ and $V_t$ are the physical volumes represented by the masks $M_0$ and $M_t$, respectively.

6. The intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage according to claim 4, characterized in that, The calculation process for the change in texture entropy $\Delta Entropy$ includes: Within the common area defined by the masks $M_0$ and $M_t$, the image block is first subjected to gray-level normalization to eliminate intensity differences, and then its gray-level co-occurrence matrix is ​​calculated. The entropy of the normalized gray-level co-occurrence matrix is ​​calculated using the formula $Entropy = -\sum_{i=1}^{N_g} \sum_{j=1}^{N_g} p(i,j) \log_2 p(i,j)$, and then the entropy change $\Delta Entropy$ is obtained.

7. The intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage according to claim 1, characterized in that, The warning rule base in the intelligent warning judgment module is dynamically updated. It is configured to receive user feedback on warning results and, based on this feedback, adaptively adjust the thresholds or logical combinations in the rules through a reinforcement learning algorithm.

8. The intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage according to claim 4, characterized in that, The early warning rule base contains one or more of the following rules: If the volume change rate (VCR) is greater than 25 percent, the warning level is "significant progress"; If the volume change rate (VCR) is greater than 15 percent and the change in texture entropy (Delta Entropy) is greater than 0.1, the warning level is "progress concern". If the volume change rate (VCR) is less than 20% and the absolute value of the average grayscale value change (Delta Mean) is greater than the preset sensitivity threshold, then the warning level is "significant absorption".

9. A smart monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage according to any one of claims 1 to 8, characterized in that, It also includes a data quality assessment module, configured to assess the signal-to-noise ratio and contrast of the registered image $I_t'$ in the neighborhood of the reference coordinate P0 before the dynamic change quantization module performs calculations; if the assessment result does not meet the preset quality standard, it will trigger data re-acquisition or prompt manual review.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the function of an intelligent monitoring and early warning system for dynamic changes in cerebrospinal fluid leakage as described in any one of claims 1 to 9.

Images