Efficacy evaluation and remote follow-up platform for ultrasound intervention treatment of mastitis

Abstract

The present application relates to the field of medical informatization and remote diagnosis and treatment technology, and discloses a curative effect evaluation and remote follow-up platform for mammitis ultrasonic interventional treatment. The platform integrates the imageological changes, physiological parameter trajectories, operation records and follow-up feedback of patients through a data processing module. A window recognition module dynamically divides curative effect observation windows and conventional observation windows according to treatment and feedback time points. A patient clustering module clusters patients based on the timing mode of physiology and images within the curative effect window, and identifies different response modes. A feature extraction module extracts typical operation features and follow-up response baselines of each patient community accordingly. A strategy generation module calculates the multidimensional matching degree by matching new patient data with the feature baselines of each community, and generates individualized follow-up strategies, thereby realizing adaptive and accurate definition of the curative effect evaluation window and individualized and directional optimization of follow-up management.

Description

Technical Field

[0001] This invention relates to the field of medical information technology and telemedicine, specifically to a platform for evaluating the efficacy and conducting remote follow-up of ultrasound-guided interventional treatment for mastitis. Background Technology

[0002] In the clinical practice and postoperative management of ultrasound-guided interventional treatment for mastitis, efficacy assessment and patient follow-up mainly rely on examinations at discrete time points and patient self-reports. Current techniques typically employ fixed time intervals for follow-up examinations, combined with assessments based on physiological parameter records from each individual visit. Follow-up plans are often uniform, static templates, lacking dynamic correlation with individual treatment events and recovery progress. This model analyzes treatment procedures, imaging changes, continuous physiological indicators, and patient feedback within disjointed timeframes, failing to establish a coherent temporal correlation between treatment behavior, biological response, and long-term outcomes.

[0003] The shortcomings of existing technologies lie primarily in the mechanical nature of the assessment window and the limited scope of data analysis. Fixed-period assessments cannot accurately capture critical time stages such as inflammation subsidence, complication occurrence, or disease recurrence after interventional treatment, leading to delayed or distorted judgments on treatment response speed and patterns. Simultaneously, discrete and discontinuous data acquisition and analysis struggle to characterize the synergistic evolution between physiological parameters and imaging features, failing to distinguish the inherent differences in treatment response rhythms and patterns among different patients. This results in a blanket treatment of the patient population, with a lack of personalized basis for the intensity, frequency, and focus of clinical follow-up, impacting long-term management efficiency and treatment quality. A technological solution is needed that can dynamically define assessment stages based on actual treatment and feedback events, and automatically identify patient response patterns by integrating multi-dimensional time-series data to support individualized efficacy assessment and follow-up decisions. Summary of the Invention

[0004] The purpose of this invention is to provide a platform for efficacy evaluation and remote follow-up of ultrasound-guided interventional treatment for mastitis, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a platform for efficacy evaluation and remote follow-up of ultrasound-guided interventional treatment for mastitis, the platform comprising:

[0006] The data processing module acquires composite treatment records generated by the patient during multiple rounds of ultrasound-guided interventional treatment. These composite treatment records include imaging change markers, physiological parameter fluctuation trajectories, detailed records of interventional procedures, and follow-up feedback records.

[0007] The window recognition module identifies the efficacy observation window and the routine observation window on the timeline based on the operation time in the intervention operation record and the feedback time in the follow-up feedback record.

[0008] The patient clustering module, within the efficacy observation window, clusters patients based on the fluctuation patterns of physiological parameter trajectories and the evolution paths of imaging change markers, forming patient groups with different efficacy response patterns.

[0009] The feature extraction module extracts the detailed features of interventional procedures for each patient group and combines them with the follow-up feedback records of all patients in the regular observation window to generate typical operation features and follow-up response baselines for each patient group.

[0010] The strategy generation module matches the composite treatment records of new patients with the typical operational characteristics and follow-up response baselines of each patient group, calculates the multidimensional matching degree of the new patient to each patient group, and determines the target follow-up strategy for the new patient based on the multidimensional matching degree.

[0011] Preferably, the step of identifying the efficacy observation window and the routine observation window on the timeline based on the operation time in the interventional procedure record and the feedback time in the follow-up feedback record specifically includes:

[0012] For each patient, extract the sequence of detailed interventional procedures and the sequence of follow-up feedback records ordered by time from the composite treatment records;

[0013] Locate the operation time corresponding to each interventional operation on the timeline, and extract a preset time interval forward and backward from the operation time as the center. Mark the interval corresponding to each operation time as the efficacy observation window.

[0014] On the timeline, the interval between two adjacent efficacy observation windows, as well as the interval before the start of the first efficacy observation window and after the end of the last efficacy observation window, are merged and marked as regular observation windows.

[0015] Preferably, within the efficacy observation window, patients are clustered based on the fluctuation patterns of physiological parameter trajectories and the evolution paths of imaging change markers to form patient groups with different efficacy response patterns, specifically including:

[0016] Within a single therapeutic observation window, the parameter values ​​and their occurrence times corresponding to all local extreme points in the patient's physiological parameter fluctuation trajectory are extracted to form a physiological parameter extreme value sequence; simultaneously, the feature vectors corresponding to all imaging change markers within a single therapeutic observation window are extracted to form an imaging feature sequence.

[0017] The mean sequence of the extreme values ​​of physiological parameters for each patient across all efficacy observation windows is calculated as the physiological parameter representation sequence for the patient; the mean sequence of the imaging feature sequences for each patient across all efficacy observation windows is calculated as the imaging feature representation sequence for the patient.

[0018] Clustering of the physiological parameter representation sequences of all patients yields primary patient groups based on physiological parameters; clustering of the imaging feature representation sequences of all patients yields primary patient groups based on imaging features.

[0019] Patients in the primary groups of physiological parameters and imaging characteristics are selected, and the overlap of members in the primary groups of patients exceeds a preset overlap threshold. Patients in each group are then merged to form a patient population.

[0020] Preferably, the step of extracting detailed interventional procedure features for each patient population and combining them with follow-up feedback records of all patients within the routine observation window to generate typical operational features and follow-up response baselines for each patient population specifically includes:

[0021] For a patient population, extract the set of operation parameters recorded by all patients in each interventional procedure record. The set of operation parameters includes energy settings, duration of action, and puncture depth.

[0022] The mean and variance of each parameter in the set of operating parameters are calculated across all records to constitute the typical operating characteristics of the patient population. The typical operating characteristics include the typical values ​​and fluctuation ranges of each parameter.

[0023] Within the regular observation window, the follow-up feedback records of all patients in the patient population are extracted. The follow-up feedback records include symptom description text and subjective comfort scores.

[0024] Keyword extraction and frequency statistics were performed on the symptom description text to obtain a high-frequency symptom word set; the mean of the subjective comfort scores was calculated to obtain the average comfort score;

[0025] The high-frequency symptom vocabulary and the average comfort score together constitute the follow-up response baseline for the patient population.

[0026] Preferably, the step of matching the new patient's composite treatment record with the typical operational characteristics and follow-up response baseline of each patient population to calculate the multidimensional matching degree of the new patient's belonging to each patient population specifically includes:

[0027] For a patient population to be matched, the detailed records of interventional procedures are extracted from the composite treatment records of new patients to obtain the set of operation parameters for the new patients.

[0028] Calculate the absolute value of the difference between each parameter in the set of operational parameters of a new patient and the typical value of the corresponding parameter in the typical operational characteristics of the patient population, and divide the absolute value of the difference by the fluctuation range of the corresponding parameter to obtain the normalized deviation of each parameter;

[0029] The normalized deviations of all parameters are weighted and summed to obtain the operational deviation index of the new patient from the patient population in the operational characteristic dimension.

[0030] Extract the follow-up feedback records of new patients within the routine observation window from their composite treatment records to obtain the symptom description text and subjective comfort score of the new patients;

[0031] Calculate the text similarity between the symptom description text of the new patient and the set of high-frequency symptom words in the follow-up response baseline of the patient population; calculate the absolute value of the score difference between the subjective comfort score of the new patient and the average comfort score in the follow-up response baseline of the patient population.

[0032] The text similarity and the absolute value of the score difference are weighted and combined to obtain the response deviation index between the new patient and the patient population in the follow-up response dimension.

[0033] The operation deviation index and the response deviation index are weighted and fused to obtain the multidimensional matching degree of the new patient belonging to the patient group, wherein the multidimensional matching degree value is negatively correlated with the probability of belonging.

[0034] Preferably, the step of determining the target follow-up strategy for new patients based on the multidimensional matching degree specifically includes:

[0035] Calculate the multidimensional matching degree of each new patient to belong to each patient population in the system to obtain a multidimensional matching degree set;

[0036] In the set of multidimensional matching degrees, find the multidimensional matching degree with the smallest value, and mark the patient population corresponding to the multidimensional matching degree with the smallest value as the potential matching population of new patients.

[0037] Obtain the set of high-frequency symptom words from the follow-up response baseline of potential matching communities;

[0038] Extract symptom description text from the follow-up feedback records of new patients within the routine observation window, and identify symptom keywords from the high-frequency symptom word set that are not included in the follow-up response baseline of the potential matching population. Mark the symptom keywords as abnormal symptom indicators of new patients.

[0039] Based on the specific content of the abnormal symptom indication set, the corresponding follow-up strategy is matched from the preset strategy mapping table as the initial strategy;

[0040] By combining the preliminary strategy with the multidimensional matching set of new patients, the follow-up frequency and follow-up content depth in the preliminary strategy are dynamically adjusted to generate the final target follow-up strategy for new patients.

[0041] Preferably, the step of extracting symptom description text from the follow-up feedback records of new patients within the routine observation window, and identifying symptom keywords from the high-frequency symptom word set that are not included in the follow-up response baseline of the potential matching population, specifically includes:

[0042] The symptom description text is segmented and stopped word filtered to obtain a list of candidate symptom keywords;

[0043] The semantic similarity of each word in the candidate symptom keyword list with each word in the high-frequency symptom word set of the follow-up response baseline of the potential matching community is calculated.

[0044] If the semantic similarity between a candidate symptom keyword and any word in the high-frequency symptom word set is lower than a preset similarity threshold, then the candidate symptom keyword is determined to be not included and added to the abnormal symptom indicator set.

[0045] Preferably, the calculation of text similarity between the symptom description text of the new patient and the high-frequency symptom word set in the follow-up response baseline of the patient population specifically includes:

[0046] The symptom description text of new patients and the set of high-frequency symptom words are represented as vectors using a word vector model;

[0047] Calculate the cosine similarity between the symptom description text vector of the new patient and the vector of each word in the high-frequency symptom word set;

[0048] The maximum value among all cosine similarities is taken as the text similarity.

[0049] Preferably, the clustering of the physiological parameter characterization sequences of all patients to obtain primary patient grouping based on physiological parameters specifically includes:

[0050] Each patient's physiological parameter sequence is treated as a data point, forming a data point set.

[0051] Calculate the Euclidean distance between any two data points in the data point set to form a distance matrix;

[0052] Based on the distance matrix, a hierarchical clustering algorithm is used to cluster the data point set. Clustering stops when the number of clusters reaches the preset number of clusters. Each final cluster is a primary group of patients based on physiological parameters.

[0053] Preferably, the physiological parameter fluctuation trajectory includes body temperature data sequences, white blood cell count data sequences, and pain visual analog scale (VAS) score data sequences collected at preset time points before and after each treatment.

[0054] Compared with the prior art, the beneficial effects of the present invention are:

[0055] This technology dynamically defines the efficacy observation window and routine observation window on a timeline based on the intervention time and follow-up feedback time, replacing the fixed-cycle assessment schedule. This allows data analysis to automatically adapt to critical response periods directly triggered by treatment events and routine monitoring periods where the condition is relatively stable, achieving real-time synchronization between the assessment timeframe and the individual treatment progress. This avoids delays or omissions in identifying key efficacy signals due to mismatches between the assessment period and the actual pathophysiological stages, enhancing the temporal accuracy and clinical relevance of efficacy assessment.

[0056] Within a dynamically defined therapeutic observation window, the continuous fluctuation patterns of physiological parameter trajectories and the sequential evolution paths of imaging changes are analyzed simultaneously, and patients are clustered accordingly. This technology achieves deep fusion and collaborative analysis of two types of heterogeneous time-series data: continuous physiological monitoring data and serial imaging features. It can automatically identify patient subgroups with different intrinsic response patterns from multi-dimensional dynamic information. The results break through the traditional approach of relying on a single time-point indicator or a single data type for patient stratification, providing objective data-driven evidence for understanding the population heterogeneity of treatment response.

[0057] Based on cluster analysis results, interventional procedure characteristics are extracted for each patient population with a common response pattern, and a follow-up response baseline is established within a regular observation window. When processing new patient data, targeted follow-up strategies are generated by quantifying the degree of matching between the patient's composite treatment records and the characteristics and baselines of each population in a multidimensional space. This allows the focus, frequency, and methods of follow-up management to be differentiated and adaptively adjusted according to the patient's response pattern category, promoting the transformation of follow-up practice from standardized protocols to personalized protocols based on response pattern classification. Attached Figure Description

[0058] Figure 1 This is a timeline diagram of the efficacy evaluation and remote follow-up platform for ultrasound-guided interventional treatment of mastitis as described in this invention;

[0059] Figure 2 A flowchart for observing window recognition;

[0060] Figure 3 A flowchart for patient clustering;

[0061] Figure 4 A multi-dimensional comparison chart of follow-up responses at different observation windows;

[0062] Figure 5 This is a semantic similarity comparison chart between symptom keywords and typical symptoms of patient groups. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0064] Please see Figure 1 This invention provides a platform for efficacy evaluation and remote follow-up of ultrasound-guided interventional treatment for mastitis. The platform includes: a data processing module that acquires composite treatment records generated during each ultrasound-guided interventional treatment, integrating multi-source data such as imaging change markers, physiological parameter fluctuation trajectories, detailed interventional operation records, and follow-up feedback records; a window recognition module that divides the time axis into two different time intervals—an efficacy observation window and a routine observation window—based on the operation time recorded in the detailed interventional operation record and the feedback time recorded in the follow-up feedback record; a patient clustering module that analyzes the fluctuation patterns of patients' physiological parameter trajectories and the evolution paths of imaging change markers within the efficacy observation window, and clusters patients into several patient groups with different efficacy response patterns; and a feature extraction module that extracts features from the detailed interventional operation record for each patient group, and combines this with the follow-up feedback records of all patients in that group within the routine observation window to generate typical operation characteristics and a follow-up response baseline for that group. The strategy generation module matches a new patient's composite treatment record with the typical operational characteristics and follow-up response baselines of existing patient groups in the system, calculates the multidimensional matching degree to which the new patient belongs to each patient group, and determines a personalized target follow-up strategy for the new patient based on the calculated multidimensional matching degree results.

[0065] In one embodiment of the present invention, see [reference] Figure 2The window recognition module identifies the implementation methods of the efficacy observation window and the regular observation window on the timeline based on the operation time in the interventional operation details and the feedback time in the follow-up feedback record. A detailed explanation is provided using a specific patient's example scenario. A patient's composite treatment record in the system includes a sequence of interventional operation details arranged chronologically, recording three ultrasound interventional treatments. The operation times in the interventional operation details sequence are 10:00 AM on May 10, 2025, 10:15 AM on May 17, 2025, and 9:45 AM on May 24, 2025. The patient's composite treatment record also includes a sequence of follow-up feedback records arranged chronologically, with feedback times including 4:30 PM on May 9, 2025, 5:00 PM on May 16, 2025, 4:45 PM on May 23, 2025, and 10:00 AM on May 31, 2025. The window recognition module locates the three operation moments mentioned above on the timeline, sets the preset time interval to 24 hours, takes the operation moment of 10:00 on May 10, 2025 as the center, extracts 24 hours forward to 10:00 on May 9, 2025, and extracts 24 hours backward to 10:00 on May 11, 2025, marking the closed interval [10:00 on May 9, 2025, 10:00 on May 11, 2025] as a therapeutic effect observation window. In practice, the window recognition module performs the same operation on the operation time of 10:15 on May 17, 2025, generating efficacy observation windows for [10:15 on May 16, 2025, and 10:15 on May 18, 2025]. The same operation is performed on the operation time of 09:45 on May 24, 2025, generating efficacy observation windows for [09:45 on May 23, 2025, and 09:45 on May 25, 2025].

[0066] In some embodiments, after the window recognition module completes the marking of all efficacy observation windows, it begins to merge the marking of regular observation windows. On the timeline, the interval before the start of the first efficacy observation window [May 9, 2025, 10:00 AM to May 11, 2025, 10:00 AM], i.e., all time before 10:00 AM on May 9, 2025, is included in the regular observation window. In specific implementations, the interval between two adjacent efficacy observation windows, such as the interval between the end time of the first efficacy observation window, May 11, 2025, 10:00 AM and the start time of the second efficacy observation window, May 16, 2025, 10:15 AM, and the interval between the end time of the second efficacy observation window, May 18, 2025, 10:15 AM and the start time of the third efficacy observation window, May 23, 2025, 9:45 AM, are all marked as part of the regular observation window. The period following the end of the last efficacy observation window [May 23, 2025, 09:45, May 25, 2025, 09:45], i.e., all time after May 25, 2025, 09:45, is also merged into the regular observation window.

[0067] Optionally, the preset time interval can be adjusted based on clinical practice. In a data comparison example, for patients with a high expected treatment response, the preset time interval might be set to 48 hours, resulting in an efficacy observation window of [operation time - 48 hours, operation time + 48 hours], which would encompass a wider range of data changes before and after treatment. For patients with a moderate treatment response or those in the later stages of recovery, the preset time interval might be set to 12 hours, resulting in an efficacy observation window of [operation time - 12 hours, operation time + 12 hours], which helps focus on the immediate effects of the treatment. Regardless of the preset time interval value, the window recognition module adheres to the principle of symmetrical truncation centered on the operation time to define the efficacy observation window, and uniformly merges and marks continuous intervals on the timeline not covered by any efficacy observation window as regular observation windows.

[0068] As can be understood, through the above process, the window recognition module structures the patient's continuous treatment and follow-up timeline into alternating efficacy observation windows and routine observation windows. The efficacy observation window closely revolves around the interventional procedure, aiming to capture the physiological and imaging changes directly induced by the treatment. The routine observation window covers the treatment interval, the pre-treatment baseline period, and the long-term follow-up period after treatment, used to assess the patient's stable recovery status and long-term efficacy. This division provides a clear and clinically meaningful temporal analysis framework for subsequent patient clustering and feature extraction. In some embodiments, the ratio of the total duration of the efficacy observation window to the total duration of the routine observation window can serve as an indicator for assessing the intensity of patient treatment. For example, in patient records with multiple treatments and short intervals, efficacy observation windows may partially overlap or be closely adjacent on the timeline, in which case the duration of the routine observation window will be compressed; while in patient records with fewer treatments and longer intervals, the routine observation window occupies the majority of the timeline. The window recognition module can automatically adapt to and handle these different situations.

[0069] In one embodiment of the present invention, see [reference] Figure 3 The patient clustering module clusters patients within the efficacy observation window based on the fluctuation patterns of physiological parameter trajectories and the evolution paths of imaging change markers. This is illustrated in detail using a simplified example scenario involving three patients. Patients A, B, and C all underwent three ultrasound-guided interventional treatments. The window recognition module generated three efficacy observation windows for each patient. In the specific implementation, for patient A's physiological parameter fluctuation trajectory within the second efficacy observation window, this trajectory includes body temperature data sequences, white blood cell count data sequences, and visual analog scale (VAS) pain data sequences collected at fixed time points before and after the treatment. The module identifies the local extreme points in the body temperature data sequence as 38.2℃ preoperatively and 36.8℃ postoperatively, and the local extreme point in the white blood cell count data sequence as 13.5 × 10⁻⁶ preoperatively. 9 / L and postoperative 9.8×10 9 / L, the local extreme points of the visual analog scale (VAS) pain score data sequence were identified as 7 points preoperatively and 4 points postoperatively. The parameter values ​​of these extreme points and their corresponding occurrence times were recorded, constituting the physiological parameter extreme value sequence of patient A in the second efficacy observation window. Simultaneously, all imaging change markers of patient A within the same efficacy observation window were extracted, such as "hypoechoic area reduced by 15%" and "blood flow signal grade decreased from grade III to grade II". Each marker was converted into a numerical feature vector, and the image feature sequence of patient A in that window was constructed in chronological order. For patients A, B, and C, the mean sequence of their respective physiological parameter extreme value sequences in all three efficacy observation windows was calculated, resulting in the physiological parameter characterization sequences of patient A, patient B, and patient C. The mean sequence of their respective image feature sequences in all three efficacy observation windows was also calculated, resulting in the image feature characterization sequences of patient A, patient B, and patient C.

[0070] In some embodiments, after calculating the representation sequences for all patients, the module clusters the physiological parameter representation sequences of all patients. The physiological parameter representation sequences of patient A, patient B, patient C, and all other patients are treated as independent data points, forming a data point set. The Euclidean distance between any two data points in the data point set is calculated using the following formula:

[0071]

[0072] in: This represents the Euclidean distance between two data points x and y. and Let x and y represent the values ​​of data point x and y in the i-th feature dimension, respectively. This represents the total number of feature dimensions. Based on the calculated distance matrix, the average linkage method in hierarchical clustering is used to cluster the data point set. When the number of clusters formed by the clustering process reaches a preset number, for example, 4 clusters, the clustering process stops. At this point, each final cluster is a primary patient group based on physiological parameters. The same process is repeated completely independently for the image feature representation sequences of all patients, i.e., calculating the distance matrix and using hierarchical clustering to cluster, to obtain primary patient groups based on image features.

[0073] Optionally, the preset number of clusters and the linking method of the hierarchical clustering algorithm are adjustable parameters. In a data comparison example, when the preset number of clusters is set to 5, the primary patient grouping based on physiological parameters may group patients with a rapid decline followed by a plateau pattern in their physiological parameter representation sequences into different groups compared to those with a slow, stepwise decline pattern. However, when the preset number of clusters is set to 3, patients with both patterns may be merged into the same primary patient group based on physiological parameters. Similarly, the results of primary patient grouping based on imaging features are also affected by the preset number of clusters. The module then filters out group pairs from the primary patient group based on physiological parameters and the primary patient group based on imaging features where the overlap exceeds a preset overlap threshold, which can be set to 70%. Suppose a primary patient group P1 based on physiological parameters contains patients A, D, and E, and a primary patient group I2 based on imaging features contains patients A, D, and F. The intersection of the members of group P1 and group I2 is patients A and D. The overlap of members is calculated by dividing the number of members in the intersection by the average number of members in the group, i.e., 2 divided by ((3+3) / 2), which is approximately 66.7%. If this value is lower than the preset overlap threshold of 70%, then group P1 and group I2 will not be merged. If the calculated overlap of members for another group reaches 80%, exceeding the preset overlap threshold, then all patients in that group will be merged to form a final patient population.

[0074] It is understandable that, through the above process, the patient clustering module integrates the dynamic physiological response patterns and imaging structural evolution paths of patients within the efficacy observation window, dividing the patient population into multiple communities with consistent internal characteristics. The physiological parameter fluctuation trajectory provides real-time, quantitative information about the body's response, while imaging change markers provide objective evidence of structural changes. The mean sequence of both characterizes the overall response trend of patients across multiple treatment cycles. In specific implementations, the use of hierarchical clustering algorithms allows grouping based on the distribution characteristics of the data itself, without needing to pre-specify the center of the group. The introduction of a preset overlap threshold ensures that the final patient communities simultaneously meet the consistency requirements of physiological response and imaging evolution, avoiding the bias that may arise from clustering based solely on single-dimensional information. In some embodiments, for large-scale patient data, the physiological parameter representation sequence and the imaging feature representation sequence can be dimensionality-reduced first, and then clustering calculations can be performed to improve computational efficiency. Optionally, the calculation of member overlap can also employ other mathematical forms, such as the Jaccard similarity coefficient. Regardless of the specific parameter settings, the core steps of the patient clustering module are always to extract the bimodal sequences within the efficacy observation window, calculate the patient-level representation sequences, perform dual clustering, and filter and merge based on overlap.

[0075] In one embodiment of the present invention, the feature extraction module extracts the interventional procedure details of the patient population and combines them with the follow-up feedback records of all patients within the conventional observation window to generate typical operation characteristics and follow-up response baselines for each patient population. This implementation is described in detail through an example scenario containing a patient population Alpha. The patient population Alpha is generated by the patient clustering module and contains five patients. The feature extraction module first extracts the set of operation parameters recorded in the interventional procedure details of all patients within the patient population Alpha. The set of operation parameters includes energy settings, duration of action, and puncture depth. For example, the set of operation parameters extracted from the detailed record of an interventional procedure for patient P001 in patient community Alpha is recorded as "Energy setting: 30 joules; Duration of action: 120 seconds; Puncture depth: 15 mm". The set of operation parameters extracted from the detailed record of an interventional procedure for patient P002 in patient community Alpha is recorded as "Energy setting: 28 joules; Duration of action: 115 seconds; Puncture depth: 14 mm". The feature extraction module traverses all the detailed records of all interventional procedures for all patients in patient community Alpha and summarizes the values ​​of all operation parameter sets. The feature extraction module then calculates the mean and variance of each parameter in the operational parameter set across all aggregated records. For the energy setting parameter, the arithmetic mean of the energy setting values ​​across all records is calculated as the typical value of the energy setting, and the variance of the energy setting values ​​across all records is calculated as a measure of the energy setting's fluctuation range. For the duration of action parameter, the arithmetic mean of the duration of action values ​​across all records is calculated as the typical value of the duration of action, and the variance of the duration of action values ​​across all records is calculated as a measure of the duration of action's fluctuation range. For the puncture depth parameter, the arithmetic mean of the puncture depth values ​​across all records is calculated as the typical value of the puncture depth, and the variance of the puncture depth values ​​across all records is calculated as a measure of the puncture depth's fluctuation range. These typical values ​​and fluctuation ranges together constitute the typical operational characteristics of the patient population Alpha, where the variance calculation formula is:

[0076]

[0077] in: Represents the variance of a certain operating parameter. This represents the total number of times this operational parameter appears in the detailed records of all interventional procedures in the Alpha patient population. This represents the specific value of the operation parameter in the i-th record. This represents the arithmetic mean of all recorded values ​​for this operation parameter.

[0078] In some embodiments, the feature extraction module extracts the follow-up feedback records of all patients in the patient community Alpha within a regular observation window. These follow-up feedback records include symptom description texts and subjective comfort scores. For example, from a follow-up feedback record of patient numbered P003 in the patient community Alpha, the extracted symptom description text is "There is a slight dull pain at the puncture site, and the breast swelling and redness have subsided", and the subjective comfort score is 7 points; from a follow-up feedback record of patient numbered P004 in the patient community Alpha, the extracted symptom description text is "The dull pain has basically disappeared, and there are hard nodules when touched", and the subjective comfort score is 8 points. The feature extraction module aggregates all the symptom description texts of all patients in the patient community Alpha within the regular observation window, performs automated keyword extraction and word frequency statistics on the aggregated symptom description texts. In the keyword extraction process, stop words such as "de", "he" are removed, and the occurrence times of medical description words such as "dull pain", "swelling and redness", "hard nodules", "fever" are counted. The words with an occurrence frequency exceeding a preset threshold, such as words with an occurrence frequency exceeding 10% in all texts, are screened out to form the high-frequency symptom word set of the patient community Alpha. At the same time, the feature extraction module calculates the arithmetic mean of all the aggregated subjective comfort scores to obtain the average comfort score of the patient community Alpha. The high-frequency symptom word set of the patient community Alpha and the average comfort score of the patient community Alpha are combined together to form the follow-up response baseline of the patient community Alpha.

[0079] Optionally, the composition of the operation parameter set is not limited to energy setting, action duration, and puncture depth. In a data comparison example, for patient communities using different ultrasound intervention techniques, the operation parameter sets recorded in their intervention operation details may be different; for the patient community Beta using the focused ultrasound ablation technique, its operation parameter set may include "acoustic power", "irradiation time", "focal depth"; for the patient community Gamma using ultrasound-guided percutaneous abscess aspiration, its operation parameter set may include "puncture needle type", "aspiration negative pressure", "irrigation fluid volume". When the feature extraction module processes different patient communities, it will calculate the corresponding typical values and fluctuation ranges according to the type of operation parameter set actually recorded in the patient community, so as to generate typical operation features adapted to different intervention techniques. It can be understood that the typical operation features quantitatively describe the general operation parameter levels and their variation degrees when a certain patient community undergoes ultrasound intervention treatment. For example, the typical operation features of the patient community Alpha may show that this community tends to receive treatment with medium energy and medium duration, and the parameter settings are relatively concentrated and the fluctuation range is small; while the typical operation features of the patient community Delta may show that the fluctuation range of its parameter settings is larger, reflecting a more individualized or exploratory treatment strategy.

[0080] In some embodiments, the generation method of the high-frequency symptom term set can be adjusted. Optionally, the preset frequency threshold can be set as absolute frequency, such as a certain symptom word appearing at least 5 times in all follow-up feedback records of the patient group, rather than a relative frequency percentage. Another optional implementation is that after extracting keywords from the symptom description text, not only is the word frequency counted, but the semantic similarity between words is further calculated using a word vector model. Words with highly similar semantics are grouped together, for example, "pain" and "pain sensation" are grouped into the same characterization word, and then frequency statistics and filtering are performed to generate a more semantically aggregated high-frequency symptom term set. The calculation of the average comfort score can also use different statistical measures, such as the median or the average after removing the highest and lowest scores, to reduce the impact of extreme score values ​​on the overall characterization. The typical operational features and follow-up response baselines independently generated by the feature extraction module for each patient group together constitute a structured profile describing the commonalities of treatment operations and follow-up responses of the group. Understandably, the set of high-frequency symptom terms in the follow-up response baseline reflects the set of symptoms most frequently reported by the patient population during the routine observation window, i.e., during non-acute treatment interventions, while the average comfort score quantifies the overall subjective feeling level of the population during the recovery period. The combination of the two provides a reference benchmark for the expected response status of patients in subsequent new patient matching and follow-up strategy generation.

[0081] In one embodiment of the present invention, the strategy generation module matches the composite treatment record of a new patient with the typical operational characteristics and follow-up response baseline of each patient population, and calculates the multidimensional matching degree of the new patient belonging to each patient population. This is illustrated in detail using an example scenario of a new patient E and a patient population Z to be matched. For patient population Z, the strategy generation module first extracts the detailed interventional operation record from the composite treatment record of the new patient E to obtain the operation parameter set of the new patient E. The operation parameter set records the operation parameters of the new patient E's most recent ultrasound interventional treatment, including an energy setting of 35 joules, an action duration of 110 seconds, and a puncture depth of 16 mm. The typical operational characteristics of patient population Z are pre-generated by the feature extraction module, wherein the typical value of the energy setting parameter is 30 joules with a fluctuation range of 4 joules, the typical value of the action duration parameter is 120 seconds with a fluctuation range of 10 seconds, and the typical value of the puncture depth parameter is 15 mm with a fluctuation range of 2 mm. The strategy generation module calculates the absolute value of the difference between each parameter in the new patient E's set of operational parameters and the typical value of the corresponding parameter in the typical operational characteristics of patient population Z. The calculation process is as follows: the absolute value of the difference for the energy setting parameter is |35-30|=5 joules, the absolute value of the difference for the duration parameter is |110-120|=10 seconds, and the absolute value of the difference for the puncture depth parameter is |16-15|=1 mm. The strategy generation module then divides the absolute value of the difference for each parameter by the fluctuation range defined for the corresponding parameter in the typical operational characteristics of patient population Z to obtain the normalized deviation for each parameter. The calculation process is as follows: the normalized deviation for the energy setting parameter is 5 / 4=1.25, the normalized deviation for the duration parameter is 10 / 10=1.00, and the normalized deviation for the puncture depth parameter is 1 / 2=0.50.

[0082] In some embodiments, the strategy generation module performs a weighted summation of the normalized deviations of all parameters. The weighting coefficients can be set according to clinical importance; for example, the weighting coefficient for the energy setting parameter can be set to 0.5, the weighting coefficient for the duration of action parameter to 0.3, and the weighting coefficient for the puncture depth parameter to 0.2. The weighted summation formula is as follows:

[0083]

[0084] in: An index representing the operational deviation of new patients from the patient population in terms of operational characteristics. Represents the number of operation parameters. The weight coefficient representing the j-th operation parameter and satisfying , This represents the normalized deviation of the j-th operational parameter. Based on the example values ​​above, this is the operational deviation index between the new patient E and the patient population Z in the operational feature dimension. The strategy generation module then extracts the follow-up feedback record from the new patient E's composite treatment record within the regular observation window, obtaining the symptom description text for the new patient E as "persistent stinging sensation in the treatment area, accompanied by slight skin warmth," and the new patient E's subjective comfort score as 5 points. The follow-up response baseline for patient population Z includes the high-frequency symptom terminology {"distending pain," "induration," "redness and swelling"} and an average comfort score of 7 points. The strategy generation module calculates the text similarity between the symptom description text of new patient E and the high-frequency symptom word set in the follow-up response baseline of patient group Z. Using a word vector model, each word in the symptom description text of new patient E and the high-frequency symptom word set of patient group Z is represented as a vector in a high-dimensional space. The cosine similarity between the symptom description text vector of new patient E and the vector of each word in the high-frequency symptom word set is calculated. Assuming that the calculated cosine similarity with "bloating and pain" is 0.35, the cosine similarity with "hard nodule" is 0.10, and the cosine similarity with "redness and swelling" is 0.60, the maximum value of all cosine similarities, 0.60, is taken as the text similarity.

[0085] Optionally, the weighting coefficients in the calculation formula for the operational deviation index can be adjusted based on clinical research data on the impact of different treatment parameters on efficacy. In a data comparison example, if the energy setting is considered to have the most critical impact on treatment effect, its weighting coefficient might be set to 0.6; while in another scenario where operational safety is more emphasized, the weighting coefficient for the puncture depth parameter might be assigned a higher value, such as 0.4. Different weighting assignments directly affect the final value of the operational deviation index, and thus affect the calculation results of the multidimensional matching degree. The strategy generation module calculates the absolute value of the difference between the subjective comfort score of new patient E and the average comfort score in the follow-up response baseline of patient group Z as |5-7|=2 points. The strategy generation module weights and combines text similarity with the absolute value of the score difference to obtain the response deviation index. For example, setting the weighting coefficient for text similarity to 0.7 and the weighting coefficient for the absolute value of the score difference to 0.3, and normalizing the score difference (e.g., dividing by the score range of 10), then the response deviation index of new patient E in the follow-up response dimension compared with that of patient group Z is... Where (1 - text similarity) is used to convert similarity into difference. The strategy generation module ultimately weights and fuses the operation deviation index and the response deviation index, for example, assigning them equal weights of 0.5, thus determining the multidimensional matching degree of new patient E belonging to patient population Z. It is understandable that the multidimensional matching score is negatively correlated with the likelihood of attribution; a higher score indicates a greater difference between the new patient's characteristics and the typical characteristics and baseline response of the target patient population, and vice versa. See Table 1.

[0086] Table 1: Calculation table of operational feature matching between new patient E and patient population Z

[0087]

[0088] In some embodiments, the word vector model used to calculate text similarity can be a pre-trained biomedical domain model to ensure accurate semantic representation of professional terms in symptom description text. Optionally, when calculating the symptom description text vector of a new patient E, the overall text vector can be generated by averaging all word vectors in the text, and then the cosine similarity can be calculated with the word vectors in the high-frequency symptom word set; alternatively, the maximum similarity between the symptom text of the new patient E and all words in the high-frequency symptom word set of patient group Z can be used as the text similarity, as shown in the example. It can be understood that the calculation of the response deviation index integrates the semantic information of the symptom description text and the numerical information of the subjective rating. The text similarity reflects the semantic distance between the symptom pattern reported by the new patient and the typical symptom pattern of the patient group, while the rating difference directly quantifies the numerical deviation of the subjective feeling. The weighted combination of the two constitutes a quantitative assessment of the follow-up response matching degree. The strategy generation module independently executes the above matching calculation process for each existing patient group in the system, thereby generating a set of multi-dimensional matching degree values ​​for the new patient E that includes the patient's belonging to each patient group.

[0089] See Figure 4 This is a multi-dimensional comparison chart of follow-up responses under different observation windows, used to show the differences in the distribution of symptom similarity, comfort score, and number of abnormal symptoms under the conventional / efficacy observation windows after ultrasound interventional treatment for mastitis. The subjective comfort score and symptom text similarity of the efficacy observation window are higher than those of the conventional observation window, and the number of abnormal symptoms is less, reflecting that the treatment response of patients in the efficacy observation window is better. The conventional observation window 1 has the most abnormal symptoms (3), and combined with its lower comfort score, it suggests that this window should focus on the patient's adverse reactions. For the conventional observation window (especially window 1), the follow-up frequency should be increased and the monitoring of abnormal symptoms should be strengthened; for the efficacy observation window, the follow-up density can be appropriately reduced to focus on consolidating the efficacy.

[0090] In one embodiment of the present invention, the strategy generation module determines the implementation of the target follow-up strategy for a new patient based on the multidimensional matching degree, which is described in detail using an example scenario of a new patient F. The strategy generation module calculates the multidimensional matching degree of the new patient F belonging to each existing patient population in the system, resulting in a set of multidimensional matching degrees. For example, the multidimensional matching degree of the new patient F belonging to patient population Gamma is 0.45, to patient population Delta is 0.80, and to patient population Epsilon is 0.60. In the set of multidimensional matching degrees, the strategy generation module finds the minimum multidimensional matching degree, which is 0.45, and marks the patient population corresponding to the minimum multidimensional matching degree, i.e., patient population Gamma, as the potential matching population of the new patient F. The strategy generation module obtains the high-frequency symptom word set in the follow-up response baseline of the potential matching population, i.e., patient population Gamma. The high-frequency symptom word set of patient population Gamma includes {“dull pain”, “tight skin”, “fatigue”}.

[0091] In some embodiments, the strategy generation module extracts symptom description text from the follow-up feedback records of new patient F within the regular observation window. The symptom description text of new patient F is "throbbing pain in the affected area, local heat, and fatigue". The strategy generation module performs word segmentation and stop word filtering on the symptom description text of new patient F. The word segmentation process yields a word list ["affected area", "have", "throbbing pain", "local", "heat", "feel", "fatigue"]. After filtering out stop words such as "have" and "feel", a candidate symptom keyword list is obtained ["affected area", "throbbing pain", "local", "heat", "fatigue"]. The strategy generation module calculates the semantic similarity between each word in the candidate symptom keyword list and each word in the high-frequency symptom word set of the follow-up response baseline of the potential matching patient population Gamma. The semantic similarity calculation is based on a pre-trained word vector model. The semantic similarity of the candidate symptom keyword "throbbing pain" with the high-frequency symptom word "dull pain" was calculated to be 0.78, with "tight skin" to be 0.15, and with "fatigue" to be 0.25. Similarly, the semantic similarity of the candidate symptom keyword "fever" with "dull pain" was 0.10, with "tight skin" to be 0.08, and with "fatigue" to be 0.05. The semantic similarity of the candidate symptom keyword "fatigue" with "fatigue" was 0.90. A preset similarity threshold of 0.75 was set. If the semantic similarity of a candidate symptom keyword with any word in the high-frequency symptom word set is lower than the threshold of 0.75, the candidate symptom keyword is considered not included. Based on the calculation results, the similarity between the candidate symptom keywords "throbbing pain" and "dull pain" is 0.78, which is higher than the threshold of 0.75, therefore they are not considered as not included. The similarity between the candidate symptom keyword "fever" and any high-frequency symptom words is lower than 0.75, therefore they are considered as not included. The similarity between the candidate symptom keywords "fatigue" and "weakness" is 0.90, which is higher than the threshold of 0.75, therefore they are not considered as not included. The strategy generation module adds the candidate symptom keyword "fever," which was determined to be not included, to the abnormal symptom indicator set of the new patient F.

[0092] In practice, the strategy generation module matches the corresponding follow-up strategy from a pre-defined strategy mapping table based on the specific content of the abnormal symptom indication set of the new patient F. This pre-defined strategy mapping table is a lookup table that associates specific abnormal symptom keywords with suggested follow-up actions. For example, the strategy mapping table defines: if the abnormal symptom indication includes "fever," the preliminary strategy is "increase the frequency of temperature monitoring and recommend a remote video consultation within 24 hours"; if the abnormal symptom indication includes "severe pain," the preliminary strategy is "immediately contact the attending physician to assess whether emergency intervention is needed." Since the abnormal symptom indication set of the new patient F includes "fever," the preliminary strategy matched by the strategy generation module from the strategy mapping table is "increase the frequency of temperature monitoring and recommend a remote video consultation within 24 hours." The strategy generation module combines the preliminary strategy with the multi-dimensional matching set of the new patient F, dynamically adjusting the follow-up frequency and depth of follow-up content in the preliminary strategy. This dynamic adjustment follows a predefined adjustment rule, which can be expressed as:

[0093]

[0094] in: This represents the adjusted follow-up frequency. This represents the baseline follow-up frequency defined in the initial strategy. It is a sensitivity adjustment coefficient. The average of all matches in the multidimensional matching set representing the new patient F. The minimum matching degree in the multidimensional matching degree set representing new patient F (i.e., the matching degree belonging to the potential matching community). Assume the baseline follow-up frequency of the initial strategy. The average of the multidimensional matching set of new patient F, once daily. for minimum value Set the sensitivity adjustment coefficient to 0.45. If the value is 0.5, then calculate the adjusted follow-up frequency. The strategy generation module may round this result to once daily or adjust it to once every 12 hours. Adjustments to the depth of follow-up content can be based on similar principles, such as increasing or decreasing the number of symptom items requiring assessment based on the dispersion of the multidimensional matching set. The strategy generation module combines the adjusted follow-up frequency and follow-up content depth to generate the final target follow-up strategy for the new patient F.

[0095] Optionally, the model and similarity threshold used for semantic similarity calculation can be adjusted based on actual application results. For example, a BERT model specifically fine-tuned based on clinical text can be used to calculate the semantic similarity between words, and the similarity threshold can be set to 0.8 to improve the strictness of the judgment. In a data comparison example, when the similarity threshold is set to 0.65, the similarity between the candidate symptom keywords "throbbing pain" and "dull pain" (0.78) is still higher than the threshold, but some words with weak semantic association may also be judged as already included, thus reducing the content of the abnormal symptom indicator set. However, when the similarity threshold is set to 0.85, the similarity between "throbbing pain" and "dull pain" (0.78) will be lower than the threshold, causing "throbbing pain" to also be added to the abnormal symptom indicator set, making the final generated target follow-up strategy more vigilant. It can be understood that by identifying the abnormal symptom indicator set, the strategy generation module can capture the key semantic differences in symptom descriptions between new patients and potential matching groups. These differences may indicate atypical recovery processes or complication risks, thereby driving the generation of personalized follow-up strategies.

[0096] In some embodiments, the sensitivity adjustment coefficient in the dynamic adjustment rules can be configured based on different clinical goals. If the clinical goal is to ensure close monitoring of high-risk patients, the sensitivity adjustment coefficient can be set to a larger value, making it more sensitive to differences in matching degree, thereby significantly improving the follow-up intensity for patients with poor matching degree. If the clinical goal is to optimize resource allocation and avoid over-follow-up, the sensitivity adjustment coefficient can be set to a smaller value. Optionally, dynamic adjustment not only considers the statistical characteristics of the matching degree set, but can also integrate the difference between the second smallest matching degree and the smallest matching degree in the multidimensional matching degree set of new patients. If the difference is very small, it indicates that the new patient has a similar matching degree with multiple groups, and its classification uncertainty is high. The follow-up strategy may need to incorporate more comprehensive assessment items to reduce uncertainty. The target follow-up strategy finally generated by the strategy generation module is an executable plan that includes specific follow-up actions, frequencies, and contents.

[0097] See Figure 5 This is a semantic similarity comparison chart between symptom keywords and typical symptoms of a patient population, used to assess the degree of match between symptoms reported by new patients and typical symptoms of a target population. Different symptom keywords only show high similarity with specific typical symptoms, reflecting strong semantic specificity of the symptoms. "Fever" is clearly an unmatched abnormal symptom and should be marked as a risk warning to trigger targeted follow-up or intervention. Routine follow-up can be used for highly matched symptoms (fatigue-weakness); for abnormal symptoms (fever), the frequency of follow-up should be increased and relevant examinations supplemented. "Throbbing pain" can be added to the semantic association set of "dull pain" to improve the coverage and accuracy of subsequent symptom matching. By comparing with similarity thresholds, unmatched abnormal symptoms such as "fever" can be directly located, providing clinicians with clear risk warning targets and avoiding the omission of potential adverse events.

[0098] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0099] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A platform for the evaluation of the efficacy of ultrasound intervention in the treatment of mastitis and remote follow-up, characterized in that, The platform comprises: a data processing module, which acquires a composite treatment record generated by a patient in a multi-round ultrasound intervention treatment process, the composite treatment record comprising an imaging change marker, a physiological parameter fluctuation track, an intervention operation detailed record, and a follow-up feedback record; a window identification module, which identifies, according to operation time in the intervention operation detailed record and feedback time in the follow-up feedback record, an efficacy observation window and a regular observation window on a time axis; a patient clustering module, which clusters patients in the efficacy observation window based on fluctuation patterns of the physiological parameter fluctuation track and evolution paths of the imaging change marker, to form patient communities with different efficacy response modes; a feature extraction module, which extracts intervention operation detailed record features of each patient community, and combines follow-up feedback records of all patients in the regular observation window to generate typical operation features and follow-up response baselines of each patient community; a strategy generation module, which matches a composite treatment record of a new patient with the typical operation features and follow-up response baselines of each patient community, calculates multi-dimensional matching degrees of the new patient belonging to each patient community, and determines a target follow-up strategy for the new patient according to the multi-dimensional matching degrees; The identification of the efficacy observation window and the regular observation window on the time axis according to the operation time in the intervention operation detailed record and the feedback time in the follow-up feedback record specifically comprises: For each patient, an intervention operation detailed record sequence and a follow-up feedback record sequence sorted by time are extracted from the composite treatment record; The operation time corresponding to each intervention operation detailed record is located on the time axis, and an interval of a preset time length is intercepted forward and backward from the operation time as the center, and the interval corresponding to each operation time is marked as an efficacy observation window; The interval between two adjacent efficacy observation windows, and the interval before the first efficacy observation window and after the last efficacy observation window, are combined and marked as a regular observation window on the time axis; The clustering of patients in the efficacy observation window based on fluctuation patterns of the physiological parameter fluctuation track and evolution paths of the imaging change marker to form patient communities with different efficacy response modes specifically comprises: In a single efficacy observation window, parameter values corresponding to all local extreme points in the physiological parameter fluctuation track of a patient and their occurrence time are extracted to form a physiological parameter extreme sequence; feature vectors corresponding to all imaging change markers in the single efficacy observation window are synchronously extracted to form an image feature sequence; The mean sequence of the physiological parameter extreme sequence of each patient in all efficacy observation windows is calculated as the physiological parameter representation sequence of the patient; the mean sequence of the image feature sequence of each patient in all efficacy observation windows is calculated as the image feature representation sequence of the patient; The physiological parameter representation sequences of all patients are clustered to obtain a patient primary grouping based on physiological parameters; the image feature representation sequences of all patients are clustered to obtain a patient primary grouping based on image features; A grouping pair in which the member coincidence degree of the patient primary grouping based on physiological parameters and the patient primary grouping based on image features exceeds a preset coincidence degree threshold is screened out, and the patients in each grouping pair are combined to form a patient community. The intervention operation detail feature of each patient colony is extracted, and the follow-up feedback record of all patients in the conventional observation window is combined to generate the typical operation feature and follow-up response baseline of each patient colony, specifically including: For a patient colony, the operation parameter set recorded by all patients in each intervention operation detail record is extracted, including energy setting, action duration, and puncture depth; The mean and variance of each parameter in the operation parameter set in all records are calculated to form the typical operation feature of the patient colony, including the typical value and fluctuation range of each parameter; Within the conventional observation window, the follow-up feedback record of all patients in the patient colony is extracted, including symptom description text and subjective comfort score; The keyword extraction and frequency statistics are performed on the symptom description text to obtain a high-frequency symptom word set, and the mean value of the subjective comfort score is calculated to obtain the average comfort score; The high-frequency symptom word set and the average comfort score together form the follow-up response baseline of the patient colony; The physiological parameter fluctuation trajectory includes the body temperature data sequence, white blood cell count data sequence, and pain visual analog scale data sequence collected at a preset time point before and after each treatment.

2. The mastitis ultrasonic interventional therapy efficacy evaluation and remote follow-up platform according to claim 1, characterized in that, The matching of the new patient's composite treatment record with the typical operation feature and follow-up response baseline of each patient colony is performed to calculate the multi-dimensional matching degree of the new patient belonging to each patient colony, specifically including: For a patient colony to be matched, the intervention operation detail record of the new patient is extracted from the new patient's composite treatment record to obtain the operation parameter set of the new patient; The absolute value of the difference between each parameter in the operation parameter set of the new patient and the typical value of the corresponding parameter in the typical operation feature of the patient colony is calculated, and the absolute value is divided by the fluctuation range of the corresponding parameter to obtain the normalized deviation degree of each parameter; The normalized deviation degrees of all parameters are weighted and summed to obtain the operation deviation index of the new patient in the operation feature dimension from the patient colony; The follow-up feedback record of the new patient within the conventional observation window is extracted from the new patient's composite treatment record to obtain the symptom description text and subjective comfort score of the new patient; The text similarity between the symptom description text of the new patient and the high-frequency symptom word set in the follow-up response baseline of the patient colony is calculated, and the absolute value of the score difference between the subjective comfort score of the new patient and the average comfort score in the follow-up response baseline of the patient colony is calculated; The text similarity and the absolute value of the score difference are weighted and combined to obtain the response deviation index of the new patient in the follow-up response dimension from the patient colony; The operation deviation index and the response deviation index are weighted and fused to obtain the multi-dimensional matching degree of the new patient belonging to the patient colony, wherein the multi-dimensional matching degree value is negatively correlated with the belonging probability.

3. The mastitis ultrasonic interventional therapy efficacy evaluation and remote follow-up platform according to claim 2, characterized in that, The target follow-up strategy of the new patient is determined according to the multi-dimensional matching degree, specifically including: The multi-dimensional matching degree of the new patient belonging to each patient colony in the system is calculated to obtain a multi-dimensional matching degree set; In the multi-dimensional matching degree set, find the multi-dimensional matching degree with the smallest value, and mark the patient community corresponding to the multi-dimensional matching degree with the smallest value as the potential matching community of the new patient; Obtain the high-frequency symptom word set in the follow-up response baseline of the potential matching community; Extract the symptom description text from the follow-up feedback record of the new patient within the regular observation window, and identify the symptom keywords not contained in the high-frequency symptom word set in the follow-up response baseline of the potential matching community, and mark the symptom keywords as abnormal symptom indications of the new patient; According to the specific content of the abnormal symptom indication set, match the corresponding follow-up strategy from the preset strategy mapping table as the preliminary strategy; Combine the preliminary strategy with the multi-dimensional matching degree set of the new patient, dynamically adjust the follow-up frequency and follow-up content depth in the preliminary strategy, and generate the finally determined target follow-up strategy of the new patient.

4. The mastitis ultrasonic interventional therapy efficacy evaluation and remote follow-up platform according to claim 3, characterized in that, The extraction of the symptom description text from the follow-up feedback record of the new patient within the regular observation window, and the identification of the symptom keywords not contained in the high-frequency symptom word set in the follow-up response baseline of the potential matching community, specifically includes: Carry out word segmentation processing and stop word filtering on the symptom description text to obtain a candidate symptom keyword list; Calculate the semantic similarity of each word in the candidate symptom keyword list with each word in the high-frequency symptom word set in the follow-up response baseline of the potential matching community; If the semantic similarity of a candidate symptom keyword with any word in the high-frequency symptom word set is lower than the preset similarity threshold, the candidate symptom keyword is determined as not contained, and is added to the abnormal symptom indication set.

5. The mastitis ultrasonic interventional therapy efficacy evaluation and remote follow-up platform according to claim 2, characterized in that, The calculation of the text similarity between the symptom description text of the new patient and the high-frequency symptom word set in the follow-up response baseline of the patient community specifically includes: Use a word vector model to represent the symptom description text of the new patient and the high-frequency symptom word set as vectors respectively; Calculate the cosine similarity of the symptom description text vector of the new patient with each word vector in the high-frequency symptom word set; Take the maximum value of all cosine similarities as the text similarity.

6. The mastitis ultrasonic interventional therapy efficacy evaluation and remote follow-up platform according to claim 1, characterized in that, The clustering of the physiological parameter representation sequence of all patients to obtain the patient primary grouping based on physiological parameters specifically includes: Take the physiological parameter representation sequence of each patient as a data point to form a data point set; Calculate the Euclidean distance between any two data points in the data point set to form a distance matrix; Based on the distance matrix, use a hierarchical clustering algorithm to cluster the data point set, and stop clustering when the number of clustering clusters reaches the preset cluster number. Each finally formed clustering cluster is a patient primary grouping based on physiological parameters.

Images