An intelligent control method and system for a cerebral hemorrhage postoperative rehabilitation training scheme

Abstract

The application discloses a kind of cerebral hemorrhage postoperative rehabilitation training scheme intelligent regulation and control method and system, through the acquisition of body movement data and nerve function score, establish recovery capacity evaluation scale, locate training weak functional area and quantify bilateral fatigue asymmetry degree, generate fatigue load evaluation spectrum and scheme adaptation margin;Fatigue load evaluation spectrum and scheme adaptation margin are implemented benefit diminishing analysis, identify continuous stagnation interval and construct gradient control index table;After splitting control index according to response time limit, threshold correction is implemented in combination with step back gradient detection, and control linkage parameter is generated by linkage integration;Benefit stage comparison is carried out to fusion control linkage parameter and fatigue load evaluation spectrum, the optimal scheme is identified by combining physical endurance state to enable timing, output rehabilitation training control instruction, realize the multidimensional, dynamic intelligent regulation and control of cerebral hemorrhage postoperative rehabilitation training scheme.

Description

Technical Field

[0001] This invention relates to the field of medical rehabilitation assistive technology, and in particular to an intelligent control method and system for postoperative rehabilitation training programs after cerebral hemorrhage. Background Technology

[0002] The degree of neurological and motor impairment in patients after cerebral hemorrhage surgery is highly heterogeneous due to individual differences. The development of rehabilitation training programs requires comprehensive consideration of the patient's real-time motor performance, neurological function recovery status, and differences in bilateral limb coordination. In current clinical practice, adjustments to rehabilitation programs mainly rely on the therapist's subjective assessment and experience, lacking systematic integration and quantitative analysis of multi-dimensional physiological data. This makes it difficult to promptly capture signals such as accumulated fatigue, functional decline, and decreased program suitability that patients exhibit during training.

[0003] At the level of dynamic regulation of training programs, existing methods generally suffer from delayed response. Adjustments to training parameters are often triggered only after the patient's functional state has significantly deteriorated, lacking the ability to predict the decline in training effectiveness. Simultaneously, existing systems fail to effectively distinguish the difference in response time between immediate execution parameters and long-term planning parameters, resulting in insufficient coordination between short-term regulation and long-term recovery goals. Furthermore, the timing of program switching lacks a dual verification mechanism based on both physical endurance and training effectiveness, easily introducing inappropriate high-intensity training programs during periods of insufficient patient physical strength, affecting the continuity and safety of the rehabilitation process. Summary of the Invention

[0004] This invention discloses an intelligent control method and system for rehabilitation training programs after cerebral hemorrhage. It aims to construct a comprehensive recovery capacity assessment framework by integrating limb movement data and neurological function scores. Combined with bilateral fatigue asymmetry quantification and data fit integrity verification, it achieves precise localization of training weaknesses and objective assessment of fatigue load. Furthermore, through diminishing returns analysis and training stagnation gradient detection, it drives the coordinated correction of immediate and long-term planning parameters. Finally, by integrating physical endurance status and cross-program benefit comparisons, it identifies the optimal timing for program activation and outputs rehabilitation training control commands, achieving intelligent control of the entire rehabilitation training program process.

[0005] The first aspect of this invention proposes an intelligent control method for postoperative rehabilitation training programs after cerebral hemorrhage, comprising the following steps:

[0006] Obtain limb movement data and neurological function scores, and perform functional correlation analysis to obtain a recovery ability assessment scale. Based on the recovery ability assessment scale, verify the completeness of the adaptation between the limb movement data and the neurological function scores to obtain the solution adaptation margin.

[0007] The recovery capacity assessment scale is used to locate the weak functional areas in training, and the weak functional areas are subjected to bilateral fatigue asymmetry quantification to generate a fatigue load assessment spectrum.

[0008] A diminishing returns analysis is performed on the fatigue load assessment spectrum and the scheme adaptation margin to generate diminishing returns markers. Continuous stagnation intervals are identified according to the diminishing returns markers to generate the urgency of scheme adjustment. A gradient control index table is constructed according to the urgency of scheme adjustment.

[0009] The gradient control index table is split into immediate control index and long-term control index according to the response time. The regression gradient detection is performed according to the long-term control index to obtain the regression quantification coefficient. The immediate control index is corrected according to the regression quantification coefficient to obtain the corrected control index. The corrected control index and the long-term control index are linked and integrated to generate control linkage parameters.

[0010] By integrating the aforementioned regulation and linkage parameters with the fatigue load assessment spectrum, the optimal activation time of the optimal solution is identified through benefit stage comparison. Based on the activation time of the optimal solution, the optimal rehabilitation training program is selected and rehabilitation training regulation instructions are output.

[0011] A second aspect of this invention provides an intelligent control system for postoperative rehabilitation training programs after cerebral hemorrhage, comprising:

[0012] The scale construction unit is used to acquire limb movement data and neurological function scores and perform functional correlation analysis to obtain a recovery ability assessment scale. Based on the recovery ability assessment scale, the limb movement data and neurological function scores are verified for completeness of adaptation to obtain a solution adaptation margin.

[0013] The assessment and analysis unit is used to locate the weak functional areas of training using the recovery ability assessment scale, and to generate a fatigue load assessment spectrum by performing bilateral fatigue asymmetry quantification on the weak functional areas of training.

[0014] The index construction unit is used to perform diminishing returns analysis on the fatigue load assessment spectrum and the scheme adaptation margin to generate diminishing returns markers, identify continuous stagnation intervals according to the diminishing returns markers to generate the urgency of scheme adjustment, and construct a gradient control index table according to the urgency of scheme adjustment.

[0015] The parameter integration unit is used to split the gradient control index table according to the response time to generate immediate control index and long-term control index, perform regression gradient detection according to the long-term control index to obtain regression quantification coefficient, perform threshold correction on the immediate control index according to the regression quantification coefficient to obtain the corrected control index, and combine the corrected control index and the long-term control index to generate control linkage parameters.

[0016] The scheme output unit is used to integrate the control linkage parameters with the fatigue load assessment spectrum to compare the benefits of different stages, identify the optimal scheme activation time, and select the optimal rehabilitation training scheme according to the optimal scheme activation time, and output rehabilitation training control instructions.

[0017] The beneficial effects of this invention are reflected in the following aspects: First, it establishes a multi-dimensional assessment system covering limb motor ability and neurological function, incorporating the patient's bilateral motor amplitude differences and fatigue tolerance into a unified quantitative framework. Combined with a data fit integrity verification mechanism, it achieves objective localization of weak functional areas and fine characterization of fatigue load distribution, providing a reliable data foundation for subsequent regulatory decisions and compensating for the shortcomings of traditional subjective assessment methods in terms of quantitative accuracy and data integrity. Second, through the combined diminishing returns analysis of fatigue load status and program fit margin, it establishes a mapping relationship between the decline trend of training effectiveness and the urgency of program adjustment, enabling advance prediction before the patient's functional status shows systematic decline. Furthermore, it constructs differentiated gradient regulation indicators based on the coverage and intensity of continuous stagnation intervals, providing objective quantitative evidence for the timing and magnitude of training program adjustments, thus improving the problem of delayed regulatory response in existing methods. Finally, the regulatory indicators were broken down according to response time and a regression quantification coefficient was introduced to implement graded threshold correction for the immediate parameters. At the same time, a linkage verification mechanism between the immediate execution parameters and the long-term planning constraints was established, which effectively coordinated the consistency between short-term regulation and long-term recovery goals. At the level of plan switching decision-making, the dual verification of cross-plan benefit intersection interval positioning and physical endurance status was integrated to ensure that the optimal plan activation time takes into account both training benefits and the patient's physical endurance, thereby improving the safety and timeliness of plan switching. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the intelligent control method for a post-cerebral hemorrhage rehabilitation training program according to the present invention.

[0019] Figure 2 This is a structural block diagram of an intelligent control system for post-cerebral hemorrhage rehabilitation training program according to the present invention. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0021] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0022] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0023] The technical solutions of the embodiments of this application will be described below.

[0024] like Figure 1 As shown, this embodiment of the invention provides an intelligent control method for postoperative rehabilitation training programs after cerebral hemorrhage, including the following steps S110-S150:

[0025] Step S110: Obtain limb movement data and neurological function scores, and perform functional correlation analysis to obtain a recovery ability assessment scale. Based on the recovery ability assessment scale, verify the completeness of the adaptation between the limb movement data and the neurological function scores to obtain the solution adaptation margin.

[0026] Specifically, limb movement data and neurological function scores were acquired. Wearable inertial sensors were fixed to key areas such as the wrist, ankle, and lower back on both the affected and unaffected sides, collecting triaxial acceleration and triaxial angular velocity signals for training movements such as upper limb flexion and extension, lower limb elevation, and trunk lateral flexion. Zero-bias calibration was performed before sensor use to eliminate systematic errors introduced by individual differences. The sampling frequency for limb movement data was set to 100Hz, and the acquisition duration covered the entire process of each training session from warm-up to completion. Multi-channel signals were low-pass filtered to remove baseline drift and motion noise, and then stored as two types of feature vectors: joint angle time series and motion velocity time series. Limb movement data were archived using a dual index of training date and patient number. Neurological function scores were calculated using a neurological function rating scale, administered by the assessor before and after each training session. The scale covers multiple functional domains, including limb reflex activity, joint stability, and motor coordination. Sub-item scores were assigned using a grading system, and standardized procedures ensured the comparability of scores across training periods. The neurological function score is assessed periodically every two weeks, and is archived as a vector of scores for each sub-item. The scores of each dimension in the vector correspond one-to-one with the sub-item numbers of the scale, and the overall mean of the vector reflects the patient's current comprehensive neurological function recovery level.

[0027] A recovery capacity assessment scale was obtained by performing functional correlation analysis between limb movement data and neurological function scores. Peak angles, mean angular velocities, and movement completion rates of each joint in the limb movement data were mapped item by item according to the functional zones of the neurological function assessment scale. Peak wrist angles corresponded to the joint stability functional zone, mean ankle angular velocities to the lower limb coordination functional zone, and upper limb flexion and extension amplitudes to the limb reflex activity functional zone. The mapping relationships were determined based on a combination of exercise physiology theory and clinical experience. Correlation analysis was performed between the scores of each sub-item of the neurological function score and the corresponding limb movement data mapping characteristics. Feature pairs with absolute correlation coefficients exceeding a set threshold were identified as valid correlation pairs. The number of valid correlation pairs determined the dimensional composition of the recovery capacity assessment scale. Statistical significance tests were used to filter out random correlation pairs to ensure the reliability of the scale dimensions. Each dimension of the recovery capacity assessment scale is supported by one valid correlation pair. The dimension score is the weighted average of the corresponding movement feature value and the scores of the neurological function score sub-items. The weights were taken as the absolute values ​​of the correlation coefficients to highlight the contribution of high-confidence correlations. Dimensions where limb motor data features and neurological function sub-item scores consistently fall below the healthy baseline are marked as initial low-energy zones in the recovery capacity assessment scale. The number and distribution pattern of initial low-energy zones reflect the degree of weakness in the patient's current overall recovery status. The recovery capacity assessment scale is output in a structured table format, with each row recording the current score, historical mean, and comparison with the healthy baseline for a functional dimension, supporting longitudinal trend tracking of multiple assessment results.

[0028] The completeness of the fit between limb movement data and neurological function scores was verified using a recovery capacity assessment scale to obtain a margin for program adaptation. The completeness verification proceeded item by item along each functional dimension of the recovery capacity assessment scale, verifying whether the completeness of limb movement data collection and the completeness of neurological function score coverage for that dimension met the minimum data requirements for training program execution. The completeness of limb movement data collection was calculated as the ratio of effective data collection sets to planned sets for each functional dimension's corresponding joint within the current training cycle. Effective data collection sets required continuous signal transmission and sensor non-detachment. A functional dimension with a data collection completeness below 80% indicated missing training data for that dimension. The completeness of neurological function score coverage was calculated as the ratio of actual completed assessments to planned assessments for each functional dimension's corresponding scale sub-item within the specified assessment cycle. A functional dimension with a coverage completeness below 70% indicated systematic data gaps in that dimension's assessment data. The scheme adaptation margin value for each functional dimension is defined as the difference between the weighted average of data acquisition completeness and coverage completeness and the minimum data requirement of the training scheme. The margin calculation formula is M = w1Rc + w2Rv - Tmin, where Rc is data acquisition completeness, Rv is coverage completeness, w1 and w2 are 0.6 and 0.4 respectively, and Tmin is the minimum adaptation threshold required for the execution of the training scheme in that functional dimension, with a value ranging from 0 to 1. Tmin, like Rc and Rv, is a dimensionless ratio. A negative margin for a functional dimension indicates that the execution data basis of the current training scheme in that dimension is insufficient to support effective adjustment. The scheme adaptation margin is output in the form of a vector composed of margin values ​​for each functional domain, with each component of the vector arranged according to the functional dimension number.

[0029] Step S120: Use the recovery capacity assessment scale to locate the weak functional areas in training, and perform bilateral fatigue asymmetry quantification on the weak functional areas to generate a fatigue load assessment spectrum.

[0030] In some embodiments, the step of using the recovery ability assessment scale to locate weak functional areas for training includes: extracting scores for each item of the recovery ability assessment scale according to functional dimensions to generate a functional score distribution table; performing continuous decline detection on the functional score distribution table across multiple assessments to generate continuous decline dimension markers; using the continuous decline dimension markers to perform cross-dimensional density superposition to generate an inflection point density distribution; and locating functional dimensions with scores below a threshold according to the inflection point density distribution to label weak functional areas for training.

[0031] Scores for each functional dimension of the Recovery Capacity Assessment Scale were extracted to generate a functional score distribution table. The current score, historical mean, and comparison with the healthy baseline for each functional dimension of the Recovery Capacity Assessment Scale were extracted item by item. The extracted results were organized into a two-dimensional table structure with functional dimensions as rows and assessment indicators as columns, forming a functional score distribution table covering all dimensions. The scores of each dimension were normalized and uniformly mapped to the 0-100 range. The normalization formula was Sn=(Sr-Smin) / (Smax-Smin)×100, where Sr is the original score, and Smin and Smax are the historical lowest and highest recorded values ​​for that dimension. Normalization eliminates the interference of differences in scoring standards across different functional domains on the distribution pattern. The functional score distribution table is arranged in ascending order of current score. Dimensions with scores below 60% of the healthy baseline are highlighted with a warning indicator. In the early rehabilitation stage after cerebral hemorrhage, due to the ongoing progression of Waller's degeneration of the corticospinal tract on the affected side, the normalized scores for the lower limb coordination and wrist stability functional domains often concentrate in the low range. The percentage deviation of scores for each dimension in the functional score distribution table is recorded synchronously. The percentage deviation reflects the relative distance between that dimension and the healthy baseline. Dimensions with a deviation of more than 40% trigger a key review flag.

[0032] For example, the step of generating a continuous decline dimension marker by performing continuous decline detection on the functional score distribution table through multiple evaluations includes: extracting scores from each functional dimension in the functional score distribution table segment by segment according to the evaluation period to generate a periodic score sequence; identifying segments in the periodic score sequence where the cross-dimensional magnitude difference continuously expands to form a difference expansion area marker; using the difference expansion area marker to perform inflection point risk classification annotation to generate an inflection point risk annotation; and generating a continuous decline dimension marker according to the inflection point risk annotation.

[0033] Scores for each functional dimension in the functional score distribution table are extracted segment by segment according to the assessment period to generate a periodic score sequence. Historical score records for each functional dimension in the functional score distribution table are arranged in ascending order by assessment timestamp. The dimension scores corresponding to each assessment period in the functional score distribution table are extracted sequentially to form a one-dimensional sequence indexed by time. The sequence length is equal to the total number of historical assessment periods for that dimension. Assessment periods with missing scores are filled using linear interpolation between adjacent periods. Missing data markers are added to the interpolation positions to distinguish them from actual observations, avoiding sequence breaks caused by missing data that could affect subsequent difference calculations. Multiple periodic score sequences are stored in parallel to form a full-dimensional score time-series matrix. Rows in the matrix correspond to functional dimensions, and columns correspond to assessment periods. Post-operative recovery rates for different functional dimensions vary significantly due to the progression of cerebral edema reduction during the hemorrhage absorption period in post-stroke patients. When the periodic score sequence for lower limb coordination monotonically declines over several consecutive assessment periods, while the upper limb strength dimension remains relatively stable due to the compensatory mechanism on the healthy side, this divergence in trends constitutes a typical input signal for identifying subsequent difference amplification areas. The difference in scores between adjacent periods in the periodic score sequence is the single-period change d(t) = S(t) - S(t-1). The accumulation of multiple consecutive negative changes indicates that this dimension has entered a continuous downward state. The overall slope of the periodic score sequence is obtained by least squares linear fitting as a summary quantitative indicator of the downward trend strength. The larger the absolute value of the slope, the faster the functional decline.

[0034] The periodic score sequence identifies segments where the cross-dimensional amplitude difference continuously widens, forming a difference widening zone marker. The difference in scores between any two functional dimensions in the same assessment period is defined as the cross-dimensional amplitude difference D(i,j,t) = S(i,t) - S(j,t). A continuous increase in the absolute value of the difference reflects an accelerated differentiation in the recovery progress of the two dimensions. The amplitude difference is calculated periodically for each pair of dimensions. When the absolute value of the difference between a pair of dimensions monotonically increases over three or more consecutive assessment periods, that period is marked as the difference widening zone for that dimension pair. In patients after cerebral hemorrhage, the demyelination process of the pyramidal tract fibers on the injured side exhibits temporal differences in different functional pathways. The difference between the lower limb coordination and upper limb strength dimensions gradually widens from a few points in early assessments, continuously increasing across multiple periods, which is a typical difference widening zone. Difference widening zones involving combinations of dimensions with faster decline in the periodic score sequence are given higher priority. Identification of widening zones requires at least three consecutive periods to exclude random fluctuations caused by assessment errors. The difference expansion zone markers for all dimension pairs are summarized along the time axis. The number of dimension pairs with difference expansion in the same evaluation period is superimposed to form the expansion density for that period. Evaluation phases with dense appearance of difference expansion zone markers indicate that multiple pairs of functional dimensions are synchronously entering a graded deterioration state.

[0035] Inflection point risk classification is generated using the difference expansion zone marker. The duration L of each expansion zone in the difference expansion zone marker is measured by the number of consecutive expansion assessment periods. The difference growth rate V is calculated by dividing the difference between the absolute value of the difference in the last period and the absolute value of the difference in the first period within the expansion zone by the duration L. The risk score Pr = α × Ln + β × Vn, where α and β are the weighting coefficients of duration and growth rate, respectively, and are taken as 0.6 and 0.4. Ln and Vn are the values ​​of L and V after normalization to the maximum values ​​of the corresponding indicators in all expansion zones. A Pr higher than 0.7 is marked as a high-risk inflection point zone, 0.3 to 0.7 is marked as a medium-risk inflection point zone, and a Pr lower than 0.3 is marked as a low-risk inflection point zone. The current score level of the corresponding functional dimension during the phase of difference expansion area marking further influences risk grading. Dimensions already at a low score level will have their risk grade increased by one level under the same expansion conditions. In the later stages of recovery from cerebral hemorrhage, due to the gradual narrowing of the neuroplasticity window, when the lower limb coordination score remains below baseline and the expansion area lasts for more than four cycles, the inflection point risk for that dimension triggers high-risk labeling. Inflection point risk annotations are associated with the start cycle, end cycle, and risk grade for each difference expansion area. When high-risk inflection point risk annotations appear densely on the timeline, it indicates that the patient's functional state has entered the accelerated differentiation phase before the inflection point. After the expansion trend of the difference expansion area marking is interrupted, segments that expand again are re-labeled with new inflection point risk annotations to avoid the previous expansion area annotations masking the new round of differentiation risk.

[0036] Continuous decline dimension markers are generated based on inflection point risk annotations. Functional dimensions involved in high-risk annotations from the inflection point risk annotations are prioritized as candidate dimensions for continuous decline. The historical periodic score sequences of these candidate dimensions are used to further verify the continuous decline condition. Dimensions with negative scores for three or more consecutive assessment periods are confirmed for inclusion in the continuous decline dimension marker set. For cases where the inflection point risk annotation level is low risk but the corresponding dimension score is consistently below baseline, a lenient criterion is applied: a score decline in more than half of several consecutive periods meets the inclusion criteria. Post-cerebral hemorrhage patients with upper limb fine motor function domain scores declining in four out of six consecutive periods under low-risk annotations are still included in the continuous decline dimension marker set to prevent the slight regression caused by the slow progression of tendon contracture from being overlooked. The duration of decline, cumulative decline magnitude, and corresponding inflection point risk annotation risk level are correlated for each dimension in the continuous decline dimension marker set, with the cumulative decline magnitude serving as the weighting basis for subsequent cross-dimensional density superposition. Before outputting the continuously declining dimensions, remove the false decline dimensions caused by scale version switching or changes in assessors. The criteria for identifying false decline are that the decline is concentrated in the period of switching and there is no continuous change trend in the scores before and after the period.

[0037] Cross-dimensional density superposition was used to generate an inflection point density distribution using continuously declining dimension markers. The initial assessment period of each dimension's decline in the continuously declining dimension markers was extracted as an inflection point event on the time axis. When multiple dimensions declined simultaneously within the same assessment period, the inflection point events were superimposed and counted on the time axis. The superposition weight was allocated according to the proportion of each dimension's cumulative decline amplitude to the sum of all declining dimension amplitudes. Weight normalization ensured that high-amplitude dimensions had a stronger response contribution to the inflection point density distribution. Cross-dimensional density superposition was performed within a sliding window with a width of three assessment periods. The weighted cumulative value of inflection point events within the window of the inflection point density distribution constituted the inflection point density value for the current time period. When patients after cerebral hemorrhage enter the rehabilitation plateau period, the residual neural conduction deficits begin to dominate functional performance after the complete resolution of cerebral edema. The three dimensions of lower limb coordination, wrist stability, and trunk control often trigger declines in the same assessment stage, and the inflection point density distribution forms a significant density peak during this time period. Periods in which the density value is consistently higher than the mean are marked as high-risk deterioration intervals. The inflection point density distribution is stored in the form of a time-series curve with the assessment period on the horizontal axis and the inflection point density value on the vertical axis. The peak and trough distribution of the time-series curve intuitively reflects the fluctuation pattern of synchronous functional deterioration events in the patient's recovery process.

[0038] Weak functional areas are identified by locating functional dimensions with scores below a threshold according to the inflection point density distribution. Within the assessment period covered by the high-risk deterioration interval in the inflection point density distribution, the current score of each functional dimension is compared with the set threshold. Dimensions with scores below the threshold and continuously declining are included in the weak functional candidate set. Dimensions in the weak functional candidate set are sorted according to the density value corresponding to the inflection point density distribution, and dimensions with higher density values ​​are prioritized as the core constituent dimensions of the weak functional areas. Post-cerebral hemorrhage patients often have significant density peaks in lower limb coordination when inflection point events are triggered simultaneously across multiple assessment periods due to damage to the ipsilateral motor cortex, and are often the first to be identified as the core dimensions of the weak functional areas. The upper limb fine motor function domain, due to the relatively intact contralateral cortical compensatory pathway, usually scores higher and is usually in the middle to lower positions of the candidate set. The weak functional areas are defined as a set of functional domains, with each functional domain in the set accompanied by a weakness level. The level is divided into three levels: mild weakness, moderate weakness, and severe weakness, based on the deviation of the current score from the healthy baseline. The newly added downward dimensions before and after the density peak of the inflection point density distribution are preferentially included in the training weak functional area to capture the functional regression that newly appears in the deterioration acceleration stage. The number of functional domains and the distribution of the degree of weakness in the training weak functional area together constitute the analysis scope and priority ranking basis for subsequent bilateral fatigue asymmetry quantification.

[0039] In some embodiments, the step of generating a fatigue load assessment spectrum by bilateral fatigue asymmetric quantification of the training-weak functional area includes: extracting bilateral limb movement amplitude time series from the training-weak functional area to generate bilateral amplitude time series groups; extracting fatigue persistence features from the bilateral amplitude time series groups to generate a fatigue persistence distribution; performing left-right asymmetric joint weighting of the bilateral amplitude time series groups and the fatigue persistence distribution to generate an asymmetric fatigue coefficient; and constructing a fatigue load assessment spectrum based on the asymmetric fatigue coefficient.

[0040] Bilateral limb amplitude time series were extracted from the weak training functional areas to generate bilateral amplitude time series groups. The functional domains covered by the weak training functional areas corresponded to the amplitude signals of the affected and healthy sides collected in the limb motion data. The upper limb functional domain corresponded to the amplitude time series of the wrist and elbow joint motion angles, and the lower limb functional domain corresponded to the amplitude time series of the ankle and knee angles. The amplitude time series of each joint was represented by the peak angle of each set of repetitions. One amplitude time series was extracted from the affected and healthy sides for each functional domain. The two time series were aligned with the same time axis to form the affected-healthy side paired time series. In patients with cerebral hemorrhage, the excitability of α motor neurons on the affected side was reduced due to pyramidal tract injury. After the lower limb coordination functional area was identified as the weak training functional area, the peak angle time series of the ankle on the affected side was significantly lower than that on the healthy side in more than ten consecutive training sets. Moreover, the difference in amplitude between the two sides did not narrow as the number of sets increased because the muscle fatigue on the affected side accumulated faster. The paired time series sets of all functional domains constitute a two-sided amplitude time series group. Each time series in the two-sided amplitude time series group is divided into three sub-segments according to the training phase: warm-up, main training, and relaxation, and saved independently to support the segmented analysis of fatigue characteristics. If there is a signal interruption in the amplitude time series corresponding to a functional domain in the two-sided amplitude time series group, the mean interpolation of the complete segments before and after it is used to complete the data and mark it with a missing measurement indicator. Time series with signal interruptions lasting more than three groups are marked as low-quality time series and their subsequent analysis weight is reduced.

[0041] Fatigue persistence features were extracted from bilateral amplitude-time series to generate fatigue persistence distributions. The mean amplitude of each amplitude sequence in the bilateral amplitude-time series was calculated for each set within the main training segment. The trend of the mean amplitude with set number reflects the dynamic process of muscle fatigue accumulation. Due to impaired motor nerve innervation and disuse atrophy shifting muscle fibers towards fast-twitch muscle fibers, the mean amplitude of the affected limb often declines rapidly after the middle of the main training segment, showing earlier fatigue onset and slower recovery compared to the healthy side. The segment in the bilateral amplitude-time series where the mean amplitude continuously decreases and the decrease exceeds a set percentage is defined as a fatigue persistence segment. Within the fatigue persistence segment, the amplitude of each set is lower than 80% of the mean of the preceding segment. The number of sets between the starting and ending sets of the fatigue persistence segment is defined as the fatigue persistence duration F. A longer fatigue persistence duration indicates a longer time the muscle maintains training at a low activation level, directly related to excessively high training intensity or insufficient muscle strength reserves. In the bilateral amplitude time series, the baseline value of the warm-up segment amplitude is used to correct the fatigue judgment threshold of the main training segment, ensuring that the difference in baseline level between different training sessions does not affect the comparability of fatigue duration. The fatigue duration of the same functional domain on the affected and healthy sides is extracted separately and arranged in parallel to form bilateral fatigue duration pairs. The bilateral fatigue duration pairs of all functional domains are summarized into a fatigue duration distribution. The fatigue duration distribution is organized in the form of a matrix with functional domains as rows and fatigue duration on the affected and healthy sides as columns. The difference in values ​​between the affected and healthy sides in each row of the matrix reflects the degree of differentiation in bilateral fatigue tolerance in each functional domain. Functional domains in the fatigue duration distribution where the fatigue duration on the affected side is significantly greater than that on the healthy side indicate a significant bilateral endurance imbalance in that functional domain.

[0042] The asymmetric fatigue coefficient was generated by weighting the bilateral amplitude time series group and the fatigue persistence distribution. The amplitude asymmetry rate Ra = (Ah - Ap) / Ah was obtained by dividing the difference in mean motor amplitude between the affected and healthy sides in the same functional domain within the bilateral amplitude time series group by the mean amplitude of the healthy side. Here, Ah represents the mean amplitude of the healthy side, and Ap represents the mean amplitude of the affected side. A larger Ra value in the bilateral amplitude time series group indicates a more significant difference in bilateral motor ability. In patients with cerebral hemorrhage, the amplitude asymmetry rate reached 0.4 when the mean peak angle of the ankle on the affected side was approximately 60% of that on the healthy side due to damage to the striatum-thalamus-cortex circuit caused by basal ganglia hemorrhage. This suggests a moderate to severe motor ability imbalance in this functional domain. The fatigue persistence asymmetry rate Rf is calculated by dividing the difference between the duration of fatigue (Fp) on the affected side and the duration (Fh) on the healthy side within the same functional domain in the fatigue persistence distribution by the value of the healthy side. A positive Rf value indicates a longer duration of fatigue on the affected side, meaning poorer endurance. Insufficient endurance on the affected side often manifests earlier than a decrease in range of motion in the early stages of functional impairment because muscle fiber atrophy primarily affects the oxidative metabolic capacity of slow-twitch muscle fibers rather than peak contractile force. The asymmetric fatigue coefficient Ca = aRa + bRf is used. For functional domains primarily focused on strength, a is set to 0.6 and b to 0.4; for functional domains primarily focused on endurance, a is set to 0.4 and b to 0.6. The weighting is determined based on the dominant clinical attributes of different functional domains.

[0043] A fatigue load assessment spectrum is constructed based on asymmetric fatigue coefficients. The coefficient values ​​of each functional domain in the asymmetric fatigue coefficient vector are mapped to the corresponding fatigue load level. Coefficient values ​​below 0.2 are classified as mild load, 0.2 to 0.5 as moderate load, and above 0.5 as severe load. Severe load functional domains are prominently highlighted in the fatigue load assessment spectrum. The fatigue load assessment spectrum is structured with the functional domains of the weak training areas as the horizontal axis and the fatigue load level as the vertical axis. Post-cerebral hemorrhage patients typically experience more severe pyramidal tract damage in the lower limbs than the upper limbs, resulting in a non-uniform distribution of lower limb coordination (severe load), wrist stability (moderate load), and trunk control (mild load). The fatigue load assessment spectrum visually presents this gradient difference, suggesting that training resources should be prioritized for high-load functional domains. The historical average of the asymmetric fatigue coefficient is included in the fatigue load assessment spectrum as a reference baseline. Functional domains with current period coefficient values ​​higher than the historical average are marked as showing a fatigue aggravation trend, while those lower than the historical average are marked as showing a fatigue relief trend. The trend direction helps rehabilitation therapists assess the current effectiveness of the training program. The fatigue load assessment spectrum superimposes bilateral imbalance direction information on the fatigue level of each functional domain to distinguish between two imbalance modes: ipsilateral overload and contralateral overcompensation. The ipsilateral overload mode is more common in patients with basal ganglia hemorrhage, while the contralateral overcompensation mode is more common in patients with superficial cortical hemorrhage. The two imbalance modes are distinguished in the fatigue load assessment spectrum by the positive and negative directions of the asymmetric fatigue coefficient. For force-dominant functional domains, where the total coefficient is not easy to be negative due to the high weight of amplitude asymmetry rate, the sign direction of the fatigue persistence asymmetry rate Rf is used as an auxiliary criterion to identify the contralateral overcompensation mode.

[0044] Step S130: Perform diminishing returns analysis on the fatigue load assessment spectrum and the scheme adaptation margin to generate diminishing returns markers. Identify continuous stagnation intervals according to the diminishing returns markers to generate the urgency of scheme adjustment. Construct a gradient control index table according to the urgency of scheme adjustment.

[0045] In some embodiments, the step of performing diminishing returns analysis on the fatigue load assessment spectrum and the scheme adaptation margin to generate diminishing returns markers includes: performing a training cycle-by-training numerical comparison between the fatigue load assessment spectrum and the scheme adaptation margin to generate a cycle comparison sequence; identifying training phases in the cycle comparison sequence where the fatigue load and the scheme adaptation margin deteriorate synchronously to generate synchronous deterioration intervals; assigning weighted decreasing coefficients to the synchronous deterioration intervals according to their coverage and intensity to generate a classification decreasing marks group; and generating diminishing returns markers based on the coverage and intensity weights of the classification decreasing marks group.

[0046] The fatigue load assessment spectrum and the protocol adaptation margin are compared numerically on a training cycle basis to generate a cycle comparison sequence. The fatigue load level and fatigue change trend of each functional domain in the fatigue load assessment spectrum are arranged along the training cycle time axis, while the margin values ​​of each functional domain in the protocol adaptation margin are simultaneously arranged along the training cycle. The two types of data are aligned cycle by cycle on the same time axis to form the basis for comparison. When the functional domain marked as aggravated in the fatigue load assessment spectrum and the corresponding functional domain margin value of the protocol adaptation margin both decrease simultaneously, the comparison result for that functional domain in that cycle is marked as co-deterioration. Co-deterioration indicates that the effectiveness of the training protocol in that functional domain is beginning to decline. In patients after cerebral hemorrhage, during the rehabilitation plateau period, the decreased efficiency of damaged axonal conduction weakens training tolerance and data acquisition stability. Fatigue load aggravation and margin reduction often occur concurrently in the same training phase, forming a typical scenario of co-deterioration. The difference between the asymmetric fatigue coefficient value of the fatigue load assessment spectrum and the functional domain margin value of the scheme adaptation margin is calculated within the same period. The larger the absolute value of the difference, the more unbalanced the deterioration of the two types of indicators. The direction of the increase in the difference reflects whether the driving source of diminishing returns is biased towards the fatigue side or the data adaptation side. The comparison results of all functional domains in each training period are arranged by dual indexes of functional domain number and training period number to form a periodic comparison sequence. The distribution density of the same-direction deterioration markers in the periodic comparison sequence matrix intuitively reflects the temporal distribution law of the decline in the effectiveness of the training scheme. Functional domains with continuously negative margins in the scheme adaptation margin trigger data quality warning markers in the periodic comparison sequence.

[0047] Synchronous deterioration intervals are generated during training phases where fatigue load and program adaptation margin deteriorate synchronously in the periodic comparison sequence. Periods in the periodic comparison sequence where the same functional domain shows signs of simultaneous deterioration in several consecutive training cycles are identified as candidate synchronous deterioration segments. The duration of a candidate segment must cover at least two consecutive training cycles to exclude interference from single abnormalities. Candidate synchronous deterioration segments with overlapping temporal functional domains are merged. The number of functional domains covered by the merged segment and its duration together define the intensity of the synchronous deterioration interval. The intensity of the synchronous deterioration interval reflects the breadth and depth of the decline in training program effectiveness. In patients with cerebral hemorrhage who enter a training adaptation bottleneck during the mid-rehabilitation period, the lower limb coordination and wrist stability functional domains often experience simultaneous exacerbation of fatigue and a decrease in margin during the same training phase. The neuropathological basis of this phenomenon lies in the fact that hemorrhage foci in the posterior limb region of the internal capsule simultaneously affect the pyramidal tract fibers projecting to the lower limb and wrist. The synchronicity of the demyelination process of these two groups of fibers leads to synergistic regression in the corresponding functional domains during the same phase, resulting in a dense distribution of simultaneous deterioration across multiple functional domains in the periodic comparison sequence during this phase. The synchronous deterioration interval is associated with the start training cycle, the end training cycle, and the list of covered functional domains. Training cycles that have not entered the synchronous deterioration interval are marked as normal benefit stages. When the duration of the synchronous deterioration interval exceeds 30% of the total training plan cycle, a system review reminder for the scheme is triggered.

[0048] Synchronous deterioration intervals were assigned weighted decreasing coefficients based on their coverage and intensity to generate classification decreasing label groups. The number of functional domains covered by the synchronous deterioration interval, normalized by dividing by the total number of functional domains covered by the training program, was used as the range factor Fr. The duration of the synchronous deterioration interval, normalized by dividing by the total number of training cycles, was used as the intensity factor Fi. The weighted decreasing coefficient Wd = uFr + vFi, where u and v are the weight coefficients for range and intensity, respectively, and are set to 0.6 and 0.4. In patients after cerebral hemorrhage, the diffuse nature of neural conduction pathway damage makes the multi-functional domain linkage deterioration more significant on the overall rehabilitation process, so the weight of the range factor is moderately higher than that of the intensity factor. Synchronous deterioration intervals were divided into three categories according to the magnitude of the weighted decreasing coefficient: severe decreasing, moderate decreasing, and mild decreasing. Severe decreasing corresponds to Wd above 0.7, moderate decreasing corresponds to 0.3 to 0.7, and mild decreasing corresponds to Wd below 0.3. These three decreasing labels together constitute the classification decreasing label group. Cases where a single functional domain deteriorates independently while other functional domains remain stable are categorized as mild decline, to distinguish them from the systemic decline pattern of multi-domain coordinated deterioration. In patients with cerebral hemorrhage, due to more severe damage to the pyramidal tract in the lower limbs than in the upper limbs, deterioration of the lower limb functional domains alone is a common early mild decline scenario. At this time, the upper limb functional domains remain temporarily stable because the transhemispheric compensatory pathway established by the contralateral hemisphere through the corpus callosum fibers is still active. The proportion of severe decline markers in the classification decline marker group throughout the total training cycle directly reflects the degree of decline in the overall effectiveness of the training program; a proportion exceeding 30% indicates a high risk of systemic failure of the program.

[0049] Diminishing returns markers are generated based on a weighted average of the coverage and intensity of categorical diminishing returns marker groups. Each category of diminishing returns marker is arranged along a training cycle timeline. The position and distribution density of severely diminishing returns markers on the timeline serve as the primary basis for generating diminishing returns markers, while moderate and mild diminishing returns markers serve as secondary references. The weighted superposition of the three types of markers ensures that diminishing returns markers are responsive to varying degrees of performance degradation. Training cycles containing severely diminishing returns markers are directly labeled as strong diminishing returns markers, cycles containing moderate diminishing returns markers are labeled as weak diminishing returns markers, and mild diminishing returns markers are upgraded to weak diminishing returns markers only when strong or weak diminishing returns markers exist in adjacent cycles. Isolated mild diminishing returns markers remain as background fluctuations and are not included in the diminishing returns markers. The density distribution of diminishing returns markers on the timeline reflects the temporal concentration of performance degradation in the training scheme, with peak density corresponding to the training phase where the need for scheme adjustment is most urgent. When patients with cerebral hemorrhage enter a plateau phase of rehabilitation, the compensatory capacity of the remaining neural pathways tends to saturate. The long-term potentiation effect of synapses in the cortical region surrounding the injury gradually weakens, and the myelination rate of newly formed axonal lateral buds slows down. The depletion of neural plasticity reserves leads to a continuous reduction in the functional gain generated by training stimuli. Strong diminishing returns markers often appear consecutively across multiple training cycles. Diminishing returns markers are output in a pairwise format of training cycle number and diminishing returns intensity level. Each marker is also associated with a corresponding categorical diminishing returns marker group to trace the source of the diminishing returns.

[0050] The urgency of adjusting the training program is determined by identifying continuous stagnation intervals based on diminishing returns markers. Diminishing returns markers are arranged along the training cycle timeline and scanned cycle by cycle. A continuous stagnation interval is defined as a period where diminishing returns markers are present for three or more consecutive training cycles. The threshold for judgment is set at least three consecutive cycles to exclude interference from occasional fluctuations in performance. Continuous periods with concentrated marker density among diminishing returns markers indicate that the patient's training program is operating in an inefficient state for a long period. In patients recovering from cerebral hemorrhage, the brain's plasticity window gradually narrows in the later stages, and the functional reorganization of the injured cerebral hemisphere shifts from a rapid adaptation phase to a slow compensatory phase. The expression level of brain-derived neurotrophic factor (BDNF), which is related to synaptic plasticity, significantly declines six months after onset. The frequency and duration of continuous stagnation intervals both show an upward trend. The duration Ls of a continuous stagnation interval and the number of functional domains covered Nd together define the stagnation intensity index Is = (Ls / Tt) × (Nd / Nd_max), where Tt is the total number of training cycles and Nd_max is the total number of functional domains covered by the training program. A higher stagnation intensity index indicates a greater necessity for program adjustment. The urgency of the plan adjustment is mapped from the stagnation intensity index. Is below 0.2 corresponds to low urgency, 0.2 to 0.5 to medium urgency, and above 0.5 to high urgency. High urgency indicates that the current training plan is no longer effective in promoting functional recovery and plan adjustment must be initiated immediately. When there is a time misalignment in the stagnation periods of different functional domains, the stagnation intensity index of each functional domain is calculated independently. The overall urgency of the plan adjustment is the weighted maximum of the urgency of each functional domain. The weights are set according to the fatigue load level of each functional domain in the fatigue load assessment spectrum, with the heavily loaded functional domain having the highest weight. The urgency of the plan adjustment is output in the form of a level value, along with a detailed list of the urgency of each functional domain.

[0051] A gradient adjustment index table is constructed based on the urgency of the adjustment plan. The urgency of plan adjustments is stratified into high, medium, and low levels, with high urgency corresponding to large, immediate adjustments, medium urgency to gradual adjustments, and low urgency to fine-tuning and maintenance. The urgency level of each functional domain in the plan adjustment urgency table maps to its adjustment priority in the gradient adjustment index table. High-urgency functional domains are listed first in the gradient adjustment index table, receiving priority in training resource allocation and adjustment magnitude. The gradient adjustment index table is organized in a two-dimensional structure with functional domains as rows and adjustment dimensions as columns. The adjustment dimensions include three types of parameters: training intensity adjustment, training frequency adjustment, and rest interval adjustment. For high-urgency functional domains, the training intensity adjustment is set to be reduced by 20% to 30% of the current intensity, the training frequency is increased by one per week, and the rest interval is extended by 10% to 15%. For medium-urgency functional domains, the adjustment magnitude of each parameter is halved. For low-urgency functional domains, the current parameters are maintained with only minor adjustments within 5%. In the mid-rehabilitation phase of cerebral hemorrhage patients, if the lower limb coordination functional domain remains under high urgency due to the continued progression of pyramidal tract Waller degeneration, a gradient regulation index can be used to systematically reduce the training intensity of this functional domain and moderately increase the rest time between training sessions to promote metabolic recovery of the neuromuscular junction on the affected side. Meanwhile, the wrist stability functional domain, due to its lower urgency, maintains its original parameters. The clinical basis for this differentiated regulation strategy is that the degree of denervation of the distal lower limb muscle groups is usually greater than that of the proximal and upper limb muscle groups. The distal lower limb muscle fibers are more prone to triggering compensatory spasticity patterns under high-intensity training due to the lack of effective motor unit recruitment. Reducing the training intensity combined with extending the rest interval helps to reduce the abnormal firing frequency of muscle spindle type Ia afferent fibers, thereby alleviating the hyperactivity of the stretch reflex.

[0052] Step S140: The gradient control index table is split into immediate control index and long-term control index according to the response time. The regression gradient detection is performed according to the long-term control index to obtain the regression quantification coefficient. The immediate control index is corrected according to the regression quantification coefficient to obtain the corrected control index. The corrected control index and the long-term control index are linked and integrated to generate control linkage parameters.

[0053] Specifically, the gradient control index table is split into immediate and long-term control indicators based on response time. The control parameters for each functional domain in the gradient control index table are divided into two categories based on response time: parameters with a response time shorter than one training cycle are classified as immediate control indicators; parameters with a response time covering three or more training cycles are classified as long-term control indicators; parameters with a response time between the two are categorized according to their direct impact on the patient's current training session, with stronger impacts classified as immediate control indicators and weaker impacts as long-term control indicators. Adjustments to training intensity and rest intervals are classified as immediate control indicators because they directly affect the load level of the current training session, while adjustments to training frequency are classified as long-term control indicators because they require rescheduling the training calendar across cycles. For patients with cerebral hemorrhage who enter a high-urgency phase in the lower limb coordination functional domain, a 20% reduction in intensity in the corresponding gradient control index table can be implemented immediately and is considered an immediate control indicator, while adjusting the frequency by increasing the number of weekly training sessions requires prior coordination of rehabilitation treatment schedules and is considered a long-term control indicator. Real-time control indicators are indexed by functional domain number and parameter type, and are associated with the original threshold and adjustment direction for direct reading in subsequent threshold correction stages. Long-term control indicators are also indexed by functional domain number and associated with the planning start period and expected execution period to support cross-period scheduling. After the split, the parameter sets of the two types of indicators have no overlap, ensuring that they are processed independently without interference in the subsequent threshold correction process.

[0054] In some embodiments, the step of performing regression gradient detection to obtain regression quantification coefficients according to the long-term regulation index includes: performing time-series scanning on the long-term regulation index to identify stagnant segments of schemes that have not triggered regulation to form a stagnant segment sequence; spontaneously identifying the stagnant segment sequence according to stagnant trigger characteristics to generate spontaneous stagnant markers; classifying and weighting the stagnant segment sequence according to the spontaneous stagnant markers to generate a weighted stagnant distribution; and determining the regression quantification coefficients based on the cumulative intensity of the weighted stagnant distribution.

[0055] A time-series scan of long-term regulatory indicators identifies stagnation periods where regulation is not triggered consecutively, forming a stagnation period sequence. The training frequency adjustment amount and planning start period for each functional domain within the long-term regulatory indicators are arranged along a time axis. Within each training period, the actual execution frequency of that functional domain is compared with the planned frequency. Consecutive periods where the actual execution frequency consistently fails to reach the planning trigger threshold are defined as stagnation periods. For functional domains where the planning start period has arrived but there is no adjustment response in execution frequency, the corresponding training period is directly marked as the stagnation start point. The stagnation start point extends forward to the period before the first trigger of regulation in execution frequency, constituting a complete stagnation period. The sum of stagnation periods for each functional domain constitutes a stagnation period sequence. In patients with cerebral hemorrhage, due to postoperative physical fluctuations or periodic troughs in the excitability of the affected cerebral hemisphere cortex leading to decreased training tolerance, the planned frequency adjustment fails to be executed for multiple periods. During this period, the planning execution rate of the long-term regulatory indicators approaches zero, forming a prolonged stagnation period. Stagnation sequences are grouped and stored according to functional domains. Records with stagnation duration exceeding three training cycles are marked as long-term stagnation. Long-term stagnation indicates that the long-term regulatory response of the functional domain has been in a non-triggered state for a long time, and will be the focus of verification in subsequent spontaneous identification.

[0056] The stall sequence is spontaneously identified based on stall trigger characteristics to generate spontaneous stall markers. The trigger characteristics for each stall segment in the sequence are extracted by tracing training records before and after the stall initiation period. Stagnation segments marked as externally triggered stalls are those where the patient actively requests a reduction in training load, the assessor determines the patient's physical condition is insufficient that day, or the patient misses training due to external matters. Identification of externally triggered stalls requires confirmation through textual records in the training log and the assessor's signature. Stagnation segments in the sequence where no such external triggers are detected are classified as spontaneous stalls in the spontaneous stall markers. Spontaneous stalls indicate that the training program itself has design flaws leading to execution inertia, rather than being caused by patient subjective factors or external interference. A correlation analysis was performed between the functional domain score changes in the stagnation initiation cycle and the preceding training cycles. If the functional domain score had been declining continuously before the stagnation and the training program had not made responsive adjustments, the spontaneous characteristics of this stagnation segment in the spontaneous stagnation marker were reinforced. In patients with cerebral hemorrhage, a stagnation segment in which the lower limb coordination functional domain score continued to decline for more than three cycles due to secondary demyelination of the pyramidal tract, and the training frequency was not adjusted, is a typical spontaneous stagnation scenario. The degree of absence of external factors, Ea, is calculated as the ratio of the number of training cycles covered by the stagnation segment in which no external triggering factors were detected to the total number of cycles in the stagnation segment. Ea is 1 when there are no external factors in all cycles and 0 when there are external factors in all cycles. The intensity of the functional score decline precursor, Dp, is calculated as the ratio of the cumulative decline in functional domain score in the three consecutive cycles before the onset of stagnation to the healthy baseline for that dimension. When the ratio exceeds 1, it is truncated to 1. The spontaneous stagnation marker is marked by adding a spontaneous confidence score Cs to each spontaneous stagnation segment. The calculation formula is Cs=g1Ea+g2Dp, where g1 and g2 are the weight coefficients of the two indicators, respectively, and are taken as 0.6 and 0.4.

[0057] For example, the step of classifying and weighting the stagnation segment sequence according to the spontaneous stagnation marker to generate a weighted stagnation distribution includes: segmenting the stagnation segment sequence according to the stagnation duration to generate a baseline stagnation curve; separating the stagnation segment sequence and the baseline stagnation curve by difference to generate a residual stagnation sequence; passively identifying the direction of the residual stagnation sequence according to the spontaneous stagnation marker to generate a passive stagnation segment set; and assigning high weights to the passive stagnation segment set to generate a weighted stagnation distribution.

[0058] A baseline stagnation curve is generated by segmenting the stagnation sequence according to its duration. The duration of each stagnation segment in the sequence is grouped by functional domain. The durations of all stagnation segments within the same functional domain are arranged chronologically to form a duration sequence, reflecting the changing trend of stagnation persistence ability in that functional domain at different training stages. The duration sequence is modeled using a piecewise linear fitting method. Fitting points are selected where the slope of the duration sequence changes significantly. The number of segments is adaptively determined based on the length and fluctuation amplitude of the duration sequence; a single-segment linear fitting is used for shorter duration sequences, while two or three-segment fitting is used for sequences containing significant trend reversals. The piecewise fitting results form a continuous piecewise linear function on the time axis, i.e., the baseline stagnation curve. The baseline stagnation curve represents the expected stagnation level under normal fatigue accumulation conditions at the current training stage. Functional domains with generally longer stagnation durations in the stagnation sequence correspond to baseline stagnation curves with larger slopes, reflecting a systematic increasing trend in the duration of stagnation for that functional domain. As patients recover from cerebral hemorrhage, the plasticity reserves of damaged neural pathways are gradually depleted, and the slope of the baseline stagnation curve in each functional domain shows a slow upward trend. The rate of increase in the slope of the baseline stagnation curve in patients with hemorrhage volume exceeding 30 ml is significantly faster than that in patients with small-volume hemorrhage after three months of recovery, suggesting that large-volume hemorrhage has a more severe impact on the depletion of neural plasticity reserves.

[0059] The residual stagnation sequence is generated by difference separation between the stagnation segment sequence and the baseline stagnation curve. The residual value is obtained by subtracting the expected value of the baseline stagnation curve at the same time point from the actual duration of each stagnation segment in the stagnation segment sequence. A positive residual value indicates that the actual stagnation duration exceeds the expected duration, while a negative residual value indicates that the actual stagnation duration is less than expected. The residual values ​​of each stagnation segment are arranged in chronological order to form the residual stagnation sequence. The residual stagnation sequence eliminates the background influence of the baseline trend and focuses on reflecting abnormal components in the stagnation duration that cannot be explained by a systematic trend. In patients with cerebral hemorrhage, the glial scar around the hemorrhage site gradually matures within four to eight weeks after onset and forms a physical barrier to surrounding neurons. During this stage, the response threshold of the motor cortex on the injured side to training stimuli is significantly increased. Even if the patient maintains the willingness to train, it is difficult to complete the planned frequency, resulting in a persistent positive residual value in the residual stagnation sequence during the glial scar maturation period. The residual is particularly high in the lower limb coordination domain due to the dense distribution of pyramidal tract fibers in the internal capsule region. The fitting accuracy of the baseline stagnation curve is evaluated using the root mean square error (RMSE). A large RMSE indicates severe fluctuations in the stagnation sequence, requiring a moderate relaxation of the subsequent passive direction recognition threshold. The relaxation is adjusted linearly according to the ratio of the RMSE to the mean of the baseline stagnation curve. Continuous segments in the residual stagnation sequence with consistently positive residual values ​​exceeding 30% of the mean of the baseline stagnation curve are marked as outlier segments. The duration of these outlier segments can serve as an indirect indicator of the remaining window width for assessing neural plasticity. A dense occurrence of outlier segments in a functional domain suggests that the training scheme for that functional domain has significantly lowered its responsiveness than expected.

[0060] Passive stagnation segment sets are generated by passively identifying the direction of residual stagnation sequences based on spontaneous stagnation markers. Stagnation segments with positive residual values ​​in the residual stagnation sequences are cross-referenced with spontaneous stagnation markers. Stagnation segments that simultaneously satisfy both positive residuals and high confidence in spontaneous stagnation are included in the passive stagnation candidate set. Stagnation segments in the passive stagnation candidate set indicate that the training program failed to actively respond to functional regression signals, and the regression continues naturally without intervention. This passive continuation is particularly concerning in post-stroke patients, as long-term functional regression without intervention may trigger learned disuse, thereby accelerating ipsilateral muscle atrophy and joint contractures. The product of the spontaneous stagnation marker confidence score Cs and the corresponding residual value of the residual stagnation sequence is defined as the passive stagnation intensity Ip. The passive stagnation intensity comprehensively reflects the degree of attribution of the stagnation program design and the severity of deviation from expectations. Candidate stagnation segments with Ip exceeding a set threshold are confirmed as passive stagnation segments. Passive stagnation segments reflect the blind spots in the response of the training program to the patient's functional regression. Stagnation segments in the residual stagnation sequence with negative residual values ​​but high confidence in spontaneous stagnation markers are classified as active stagnation. Active stagnation indicates that the training scheme has compressed the actual stagnation duration to a lower level than expected through timely adjustments at this stage. Direction recognition excludes active stagnation from the passive stagnation segment set. The segments in the passive stagnation segment set are arranged in descending order of passive stagnation intensity. When all stagnation segments in the spontaneous stagnation markers are active stagnations, the passive stagnation segment set is empty, indicating that the scheme's response mechanism is operating normally.

[0061] A weighted stagnation distribution is generated by assigning high weights to the passive stagnation segment set. The passive stagnation intensity of each segment in the passive stagnation segment set is normalized and used as the high-weight base value. The normalization formula is Wb=Ip / Imax, where Ip is the passive stagnation intensity of the segment and Imax is the maximum passive stagnation intensity in the passive stagnation segment set. The final weighting coefficient for each segment is Wf=Wh×Wb, where Wh is the high-weight baseline value set to 1.5. The final weighting coefficient ensures that segments with high passive stagnation intensity have a dominant contribution in the weighted stagnation distribution. Externally triggered stagnation and active stagnation that are not included in the passive stagnation segment set in the stagnation segment sequence are both weighted with a standard weight of 1.0. The difference between the high weight and the standard weight reflects the prominent importance of passive stagnation in the evaluation of the scheme's benefits. The weighted stagnation intensities of the two types of stagnation segments are allocated along the training cycle time axis. The stagnation density value for each cycle is obtained by summing the weighted intensities of all stagnation segments covering that cycle. Post-cerebral hemorrhage patients exhibit varying rates of neuroplasticity regression in different brain regions. The pyramidal tract fibers of the lower limbs are highly sensitive to hemorrhage damage due to their concentrated distribution in the posterior limb of the internal capsule. Passive stagnation in the lower limb coordination domain often appears earlier and is more intense than in the upper limb domain. The weighting mechanism makes the contribution of passive stagnation in the lower limb domain to the peak of the weighted stagnation distribution more prominent. The weighted stagnation distribution forms a significant peak during the training phase when stagnation in the lower limb domain is concentrated and explosive. The peak of the weighted stagnation distribution time-series curve is driven by the segment with the highest concentrated intensity of passive stagnation segments. The training phase where the peak occurs corresponds to the period with the most severe blind spot in the scheme's response. When the set of passive stagnation segments is empty, the weighted stagnation distribution degenerates into a uniform weighted distribution.

[0062] The regression quantification coefficient is determined based on the cumulative intensity of the weighted stagnation distribution. The cumulative stagnation intensity *St* is obtained by summing the stagnation density values ​​of all training cycles in the weighted stagnation distribution time-series curve. Dividing the cumulative stagnation intensity by the total number of training cycles yields the average stagnation density *Sa*, which reflects the average severity of stagnation throughout the entire training cycle. Training cycles with a stagnation density value exceeding twice the mean in the weighted stagnation distribution are marked as high-density stagnation cycles. The proportion of high-density stagnation cycles to the total number of cycles is defined as the stagnation concentration *Sc*. Stagnation concentration reflects the degree of clustering of regression events on the time axis; high concentration indicates that functional regression is characterized by phased outbreaks rather than a uniform distribution. The regression quantification coefficient is jointly determined by the average stagnation density and the stagnation concentration, calculated using the formula Q = Sa × (1 + Sc). A larger Q indicates a more severe regression and a greater threshold correction magnitude for immediate control indicators. In the weighted stagnation distribution, the cumulative stagnation intensity of each functional domain component is calculated independently. The regression quantification coefficient of the functional domain component reflects the degree of regression of that functional domain. The overall value of the regression quantification coefficient is output synchronously with the values ​​of each functional domain component. The overall value is used to determine the direction of global threshold correction, and the component values ​​are used to calculate the differential correction magnitude of each functional domain. The historical mean of the weighted stagnation distribution is synchronously added to the regression quantification coefficient output to support cross-cycle regression trend comparison.

[0063] In some embodiments, the step of applying a threshold correction to the immediate control index according to the regression quantization coefficient to obtain a corrected control index includes: performing piecewise linearization on the regression quantization coefficient to generate a mild regression segment and a sharp regression segment; performing a graded step size correction on the mild regression segment and the sharp regression segment to generate a piecewise correction sequence; superimposing the piecewise correction sequence onto the original threshold of the immediate control index to generate a corrected threshold; and using the corrected threshold to update the immediate control index with over-limit constraints to obtain the corrected control index.

[0064] The regression quantification coefficient is segmented and linearized to generate mild regression and abrupt regression segments. The numerical range of the regression quantification coefficient Q is defined by segment boundaries based on clinical rehabilitation experience: Q below 0.3 is defined as the mild regression segment, Q above 0.6 as the abrupt regression segment, and the range between 0.3 and 0.6 is defined as the moderate regression transition segment. The transition segment is adjusted in subsequent grading using linear interpolation, with the interpolation formula being dm = dl + (dh - dl) × (Q - Ql) / (Qh - Ql), where dl is the adjustment step size for the mild regression segment, dh is the adjustment step size for the abrupt regression segment, Ql is the upper bound of the mild regression segment, and Qh is the lower bound of the abrupt regression segment. In the early stages of rehabilitation, patients with cerebral hemorrhage often experience slight stagnation due to insufficient training adaptation, and their regression quantification coefficient typically falls into the mild regression segment. At this stage, the cerebral cortex is still within the active window of functional reorganization, with sufficient neural plasticity reserves; immediate adjustment of the indicators only requires minor adjustments to restore performance. When the regression quantification coefficient enters the rapid regression phase, it indicates that the program stagnation has severely impacted the functional recovery process, and the efficiency of functional improvement generated by training stimulation decreases significantly as the density of glial scar tissue deepens. The boundary thresholds between the mild and rapid regression phases are set independently based on the differences in regression sensitivity of different functional domains. For endurance-dominated functional domains, due to larger fatigue fluctuations during normal training, the starting threshold of the rapid regression phase of the regression quantification coefficient is moderately increased to 0.7. For strength-dominated functional domains, due to relatively stable fatigue fluctuations, the standard starting threshold of 0.6 is maintained to avoid over-correction triggered by normal fatigue fluctuations.

[0065] The mild regression phase and the rapid regression phase are divided into phases by a graded step size adjustment to generate a segmented adjustment sequence. The adjustment step size dl corresponding to the mild regression phase is set to 5% to 10% of the original threshold of the immediate control index. The small step size adjustment ensures that the threshold adjustment in the mild regression phase does not cause drastic fluctuations in training parameters. The adjustment direction is to lower the training intensity threshold to reduce the execution threshold of the current training session and help patients maintain a basic training frequency during the mild stagnation period. The closer Q is to 0.3 in the mild regression phase, the closer the dl value is to 10% to enhance the adjustment response. The correction step size dh for the abrupt regression phase is set to 20% to 30% of the original threshold of the immediate control indicator. This large step size correction ensures that the threshold adjustment during the abrupt regression phase is sufficient to break the execution deadlock. The correction direction for the abrupt regression phase includes not only lowering the intensity threshold but also raising the rest interval threshold. This bidirectional correction simultaneously reduces the intensity of the current training load and extends the recovery time. This is particularly crucial for post-stroke patients because excessive fatigue may induce abnormally high muscle tone on the affected side, triggering a spastic pattern and exacerbating motor dysfunction. Extending the rest interval between sets helps restore the normal firing frequency of muscle spindle afferent signals, thereby reducing the risk of hyperexcitatory stretch reflexes. The correction step size and correction direction corresponding to each training cycle are combined to form correction instructions, arranged chronologically to constitute a segmented correction sequence. The regression quantification coefficient for the cycle falling into the transition phase is taken between the step size of the mild regression phase and the step size of the abrupt regression phase using the aforementioned linear interpolation formula. The interpolation result is included in the corresponding position of the segmented correction sequence.

[0066] Correction thresholds are generated by superimposing the segmented correction sequence onto the original thresholds of the immediate control index. The original training intensity threshold Hs and the original rest interval threshold Hr for each functional domain in the immediate control index are extracted as baseline values. The correction step size for the corresponding training cycle in the segmented correction sequence is applied to the baseline values. The training intensity correction threshold Hs' = Hs × (1 - ds), and the rest interval correction threshold Hr' = Hr × (1 + dr), where ds is the intensity down-adjustment ratio for that cycle and dr is the interval up-adjustment ratio, both determined by the corresponding step size of the segmented correction sequence. The superposition correction of the original thresholds for each functional domain is performed independently. The lower limb coordination functional domain typically receives a larger correction step size due to its higher regression quantification coefficient compared to the upper limb functional domain. This difference in correction amplitude between functional domains is consistent with the spatial distribution characteristics of pyramidal tract fibers in the internal capsule region. The initial correction thresholds are checked for boundary constraints; the training intensity correction threshold must not be lower than the patient's safe execution lower limit, which is determined based on the minimum stimulus equivalent required for the affected muscle group to maintain basic tension in the immediate control index. When multiple consecutive cycles in a segmented correction sequence have a rapidly decreasing step size, the cumulative reduction of the correction threshold is constrained by a total reduction limit of 30%. This prevents the training intensity from falling below the safety boundary due to continuous large step size corrections. After the cumulative reduction ratio reaches the upper limit, the correction amount of subsequent cycles is automatically truncated to zero until the step-down quantization coefficient falls back.

[0067] The modified control index is obtained by updating the real-time control index using modified thresholds to address over-limit constraints. The training intensity threshold and rest interval threshold for each functional domain in the modified threshold are checked against the safety upper and lower limits in the original records of the real-time control index for over-limit checks. If the training intensity threshold is lower than the safety lower limit, it is forcibly raised to the safety lower limit; if the rest interval threshold exceeds the safety upper limit, it is forcibly truncated to the safety upper limit. An over-limit truncation operation is marked with an over-limit flag to indicate that the regression quantification coefficient of that functional domain has exceeded the correctable range. The safety upper and lower limits are set according to the recommended training load range for different recovery stages in the post-cerebral hemorrhage rehabilitation medicine guidelines. The safety lower limit is set relatively high in the early post-acute phase to ensure sufficient training stimulation to promote neural plasticity, and moderately lowered in the mid-to-late recovery phase to accommodate individual differences in patient exercise tolerance. Functional domains that do not trigger over-limit truncation have their original thresholds directly replaced by the modified thresholds; functional domains that trigger over-limit truncation are updated with the truncated boundary values. After both types of update operations are completed, the real-time control index is fully refreshed to form the modified control index. When the threshold cutoff rate exceeds 50% of the total number of functional domains, it indicates that the current regression quantification coefficient is too high. In this case, the adjustment indicator is marked with an overall warning and it is recommended to review the long-term adjustment plan. At the same time, the details of the direction of each functional domain exceeding the limit are also output to support the rehabilitation therapist to make targeted adjustments to the frequency planning in the long-term adjustment indicator.

[0068] The system integrates modified and long-term regulatory indicators to generate regulatory linkage parameters. The modified training intensity threshold and rest interval threshold for each functional domain in the modified regulatory indicators are extracted as immediate execution parameters, while the frequency adjustment plan for each functional domain in the long-term regulatory indicators is extracted as long-term planning constraints. These two types of parameters are aligned item by item along the functional domain dimension before entering the linkage integration process. The linkage integration first performs consistency verification on the frequency plans of the functional domains that trigger limit-crossing flags in the modified regulatory indicators and the corresponding functional domains in the long-term regulatory indicators. Limit-crossing flags indicate that the modified regulatory indicators have reached the safety boundary. If the long-term regulatory indicators still maintain a high training frequency plan, there is a target conflict between the two types of parameters. The consistency verification identifies such conflicts and lowers the frequency plan of the long-term regulatory indicators for the conflicting functional domains to a level matching the safety margin of the modified regulatory indicators. Post-cerebral hemorrhage patients often face a contradiction between insufficient immediate tolerance and the demand for long-term training volume during functional recovery. Due to impaired local cerebral blood flow autoregulation after hemorrhage, the affected cerebral hemisphere experiences increased cerebral perfusion pressure fluctuations on high-frequency training days. Linkage integration alleviates this contradiction by coordinating the balance between single-session training load and weekly training frequency. The control linkage parameters integrate the immediate execution parameters and long-term planning constraints using the functional domain number as an index. The linkage verification status is divided into two states: consistent pass and limit adjustment. The limit adjustment status indicates that the long-term planning of this functional domain has been passively reduced due to the immediate parameter exceeding the limit. After the linkage verification of the control linkage parameters covers all functional domains, it is output in the form of a structured vector.

[0069] Step S150: Integrate the control linkage parameters and fatigue load assessment spectrum to compare the benefits of different stages, identify the optimal activation time of the optimal solution, and select the optimal rehabilitation training program according to the optimal activation time, and output rehabilitation training control instructions.

[0070] In some embodiments, the step of integrating the regulation linkage parameters with the fatigue load assessment spectrum to identify the optimal activation timing for the optimal solution through benefit stage comparison includes: mapping the regulation linkage parameters to the fatigue load assessment spectrum to generate a dynamic evaluation table of training benefits; locating cross-intervals of benefits across the training benefits dynamic evaluation table to generate cross-interval groups; extracting the physical endurance threshold for the corresponding time period in the fatigue load assessment spectrum using the cross-interval groups, filtering out periods of insufficient physical endurance to generate a valid activation candidate table; and determining the optimal activation timing based on the valid activation candidate table.

[0071] The regulatory linkage parameters are mapped to the fatigue load assessment spectrum to generate a dynamic evaluation table of training benefits. The immediate execution parameters and long-term planning constraints of each functional domain in the regulatory linkage parameters are extracted and aligned according to the functional domain number of the fatigue load assessment spectrum, ensuring a one-to-one correspondence between the two types of data in the functional domain dimension. The fatigue change trend of each functional domain in the fatigue load assessment spectrum is jointly evaluated with the linkage verification status of the corresponding functional domain in the regulatory linkage parameters. Functional domains in the dynamic evaluation table showing a fatigue aggravation trend and linkage verification triggering the limit are marked as dual-risk functional domains. Dual-risk functional domains indicate that the functional domain faces the dual pressure of fatigue accumulation and regulatory parameter limit. In patients after cerebral hemorrhage surgery, due to the increased response threshold of the ipsilateral motor cortex to training stimuli, dual-risk functional domains occur more frequently in patients with larger hemorrhage lesions. Load coordination is calculated by inverting the absolute value of the difference between the normalized value of the immediate execution parameter in the control linkage parameters and the asymmetric fatigue coefficient in the fatigue load assessment spectrum. The smaller the difference, the higher the coordination. When the lower limb coordination functional domain has a high normalized value of the immediate execution parameter and a high asymmetric fatigue coefficient due to delayed recovery of distal limb muscle strength after pyramidal tract injury, the load coordination score is significantly low, indicating that the training intensity needs to be further reduced. The training effectiveness dynamic assessment table is organized in a two-dimensional structure with functional domains as rows and assessment dimensions as columns. The assessment dimensions include load coordination, fatigue change trend, and linkage verification status. The scores of each dimension in a row are weighted by a weight of 0.5 for load coordination, 0.3 for fatigue change trend, and 0.2 for linkage verification status to form the comprehensive benefit score for that functional domain. The training effectiveness dynamic assessment table is arranged in ascending order of comprehensive benefit score, with low-efficiency functional domains concentrated at the beginning of the table to facilitate rehabilitation therapists to quickly locate and adjust the focus.

[0072] Cross-program benefit cross-interval groups are generated by locating cross-intervals in the dynamic evaluation table of training effectiveness. The comprehensive benefit scores for each functional domain in the dynamic evaluation table are arranged along the training cycle timeline to form the benefit time-series curve of the current program. Pre-set candidate training programs in the rehabilitation program database generate their own benefit time-series curves according to the same functional domain structure. The benefit time-series curves of the current program and each candidate program are compared on the timeline. When the comprehensive benefit score of a candidate program first exceeds that of the current program in a certain training cycle, that time point constitutes a positive crossover point. A positive crossover point indicates that the candidate program is more suitable for the patient's current functional state after that time point. When a patient with cerebral hemorrhage experiences a phased progress in brain tissue functional reorganization and their functional state begins to improve after the rehabilitation plateau period, the comprehensive benefit score of the advanced training program often surpasses that of the current maintenance program at this stage, forming a positive crossover. This stage corresponds to the active period of synaptic remodeling and axonal bud growth in the injured cerebral hemisphere, and the patient's tolerance and responsiveness to higher-intensity training stimuli increase simultaneously. A positive crossover point extending backwards to where the two benefit curves intersect again or reaches the end of the training plan is defined as a crossover interval and included in a crossover interval group. The average benefit advantage within each crossover interval in the crossover interval group is calculated by the average difference between the comprehensive benefit scores of the candidate solution and the current solution. The average benefit advantage reflects the extent of the candidate solution's sustained improvement relative to the current solution during that period. Crossover intervals of all solution pairs are integrated into crossover interval groups according to functional domain and time axis. Crossover intervals with larger benefit advantages and longer durations are prioritized and marked as key candidate activation windows.

[0073] Cross-interval groups were used to extract physical endurance thresholds for corresponding time periods in the fatigue load assessment spectrum, and periods of insufficient physical endurance were screened to generate a valid candidate list for activation. The time range of each cross-interval in the cross-interval group corresponds to a training cycle in the fatigue load assessment spectrum, and the fatigue load level and asymmetric fatigue coefficient of each functional domain within that time period were extracted as the basis for physical endurance assessment. For training cycles with a functional domain fatigue load level of severe, the physical endurance threshold was set to 70% of the current training intensity; for moderate fatigue, 80%; and for mild fatigue, 95%. When the fatigue trend in the corresponding time period of a cross-interval marked as a key candidate activation window is easing, it indicates that the patient's physical condition is in a recovery phase, and the physical endurance threshold is appropriately increased by 5%. Cross-intervals with physical endurance thresholds higher than the set minimum execution standard were marked as having sufficient physical endurance, and those lower than the standard were marked as having insufficient physical endurance. Cross-intervals with insufficient physical endurance were removed from the cross-interval group to ensure that the protocol switching candidates only retain time periods where the patient's physical condition is sufficient to bear the execution load of the new protocol. In the early stages of rehabilitation after cerebral hemorrhage, patients often have a low baseline physical fitness due to systemic deadaptation and postoperative nutritional depletion. Screening for insufficient physical strength at this stage typically removes a large number of candidate intervals, with the removal rate gradually decreasing as rehabilitation progresses. The remaining cross-intervals after screening are ranked from highest to lowest based on their benefit advantage. Cross-intervals with sufficient physical strength and a decreasing fatigue trend possess both benefit advantage and physical recovery, thus receiving higher priority. Each candidate interval in the effective activation candidate table is accompanied by an expected benefit score, E = Qe × Pt, where Qe is the average benefit advantage of the corresponding cross-interval, and Pt is the physical fitness threshold for the corresponding time period. A higher product value indicates that both benefit improvement and physical fitness support are at a high level. The effective activation candidate table is output in descending order of expected benefit score.

[0074] The optimal activation timing is determined based on an effective activation candidate list. The candidate interval with the highest expected benefit score in the effective activation candidate list is the first choice for optimal activation. The starting training cycle of the first choice candidate interval is defined as the recommended execution time point. The time distance between the recommended time point and the current training cycle reflects the urgency of switching the plan. If the time distance between the first choice candidate interval and the current training cycle exceeds three cycles, the candidate intervals with the second highest expected benefit scores are sequentially searched from the effective activation candidate list. The candidate interval with the closest time distance and an expected benefit score higher than the baseline threshold is selected as the optimal activation timing, avoiding excessively delayed recommendations that could lead to prolonged periods of inefficient training for the patient. Post-stroke rehabilitation patients have a limited window of opportunity. Neuroplasticity is most active within three to six months after the onset of the injury, and the rate of synaptic remodeling in the penumbra region surrounding the injury is significantly higher during this window than in subsequent stages. Delayed plan switching will directly lead to irreversible loss of training benefits during this critical window. Determining the optimal activation timing also requires stability verification. The benefit advantage of the candidate interval in the effective activation candidate list must remain stable in adjacent evaluation periods. It can only be confirmed if the cross-condition is met for two or more consecutive periods. High benefit cross-conditions that occur sporadically in a single evaluation do not meet the stability requirements and are therefore excluded. The optimal activation timing is associated with the recommended execution time point, the corresponding candidate solution number, and the expected return score. The top three candidate intervals in the effective activation candidate list are simultaneously output as alternative activation timings.

[0075] The optimal rehabilitation training program is selected and controlled according to the optimal program activation timing. The distance between the recommended execution time and the current training cycle determines the program switching execution mode. If the distance is zero or one cycle, it enters immediate switching mode; if the distance exceeds one cycle, it enters preparatory switching mode. In immediate switching mode, the candidate program number corresponding to the optimal program activation timing is directly activated as the execution program. The immediate execution parameters of each functional domain in the control linkage parameters are updated to the training parameters corresponding to the candidate programs. The updated content covers the training intensity target value, rest interval setting value, and single training set arrangement. The rehabilitation training control command sends all updated parameters to the training execution terminal in a functional domain-by-functional-domain package format. In preparatory switching mode, the rehabilitation training control command first outputs transitional training parameters. The transitional parameters linearly and gradually change between the current program parameters and the candidate program parameters over a periodic period. The gradient slope is set according to the expected benefit score of the optimal program activation timing. When the expected benefit score is high, the gradient period is shortened to accelerate the transition to the candidate program; when the expected benefit score is low, the gradient period is lengthened to reduce the adaptive impact of parameter mutations on the patient. After the rehabilitation training control instructions are executed, the response changes of the fatigue load assessment spectrum and the dynamic evaluation table of training effectiveness are continuously monitored. If the comprehensive effectiveness score does not improve for two consecutive training cycles after the switch, a rollback mechanism is triggered. The rollback mechanism restores the training parameters to the state before the switch and extracts alternative activation opportunities from the effective activation candidate list for reassessment. The final output of the rehabilitation training control instructions includes the target program number, training parameters for each functional domain, recommended execution time points, and transition cycle arrangements. The output format is compatible with the instruction parsing protocol of the training execution terminal to achieve end-to-end automated program deployment.

[0076] This is an intelligent control method for a post-cerebral hemorrhage rehabilitation training program, corresponding to the above-described method embodiments, to achieve the corresponding functional and technical effects. See also... Figure 2 , Figure 2 This application provides a structural block diagram of an intelligent control system 200 for post-cerebral hemorrhage rehabilitation training, comprising:

[0077] The scale construction unit 201 is used to acquire limb movement data and neurological function scores and perform functional correlation analysis to obtain a recovery ability assessment scale. Based on the recovery ability assessment scale, the limb movement data and the neurological function scores are verified for completeness of adaptation to obtain a solution adaptation margin.

[0078] The assessment and analysis unit 202 is used to locate the weak functional area of ​​training using the recovery ability assessment scale, and to generate a fatigue load assessment spectrum by performing bilateral fatigue asymmetry quantification on the weak functional area of ​​training.

[0079] The index construction unit 203 is used to perform diminishing returns analysis on the fatigue load assessment spectrum and the scheme adaptation margin to generate diminishing returns markers, identify continuous stagnation intervals according to the diminishing returns markers to generate the urgency of scheme adjustment, and construct a gradient control index table according to the urgency of scheme adjustment.

[0080] The parameter integration unit 204 is used to split the gradient control index table according to the response time to generate immediate control index and long-term control index, perform regression gradient detection according to the long-term control index to obtain regression quantification coefficient, perform threshold correction on the immediate control index according to the regression quantification coefficient to obtain the corrected control index, and combine the corrected control index and the long-term control index to generate control linkage parameters.

[0081] The scheme output unit 205 is used to integrate the control linkage parameters with the fatigue load assessment spectrum to identify the optimal scheme activation time by comparing the benefit stages, and to select the optimal rehabilitation training scheme and output rehabilitation training control instructions according to the optimal scheme activation time.

[0082] The aforementioned intelligent control system 200 for post-cerebral hemorrhage rehabilitation training can implement the intelligent control method for post-cerebral hemorrhage rehabilitation training in the above-described method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining contents of this application embodiment can be referred to the contents of the above method embodiments, and will not be repeated in this embodiment.

[0083] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for intelligently controlling a rehabilitation training program after cerebral hemorrhage, characterized in that, include: Obtain limb movement data and neurological function scores, and perform functional correlation analysis to obtain a recovery ability assessment scale. Based on the recovery ability assessment scale, verify the completeness of the adaptation between the limb movement data and the neurological function scores to obtain the solution adaptation margin. The recovery capacity assessment scale is used to locate the weak functional areas in training, and the weak functional areas are subjected to bilateral fatigue asymmetry quantification to generate a fatigue load assessment spectrum. A diminishing returns analysis is performed on the fatigue load assessment spectrum and the scheme adaptation margin to generate diminishing returns markers. Continuous stagnation intervals are identified according to the diminishing returns markers to generate the urgency of scheme adjustment. A gradient control index table is constructed according to the urgency of scheme adjustment. The gradient control index table is split into immediate control index and long-term control index according to the response time. The regression gradient detection is performed according to the long-term control index to obtain the regression quantification coefficient. The immediate control index is corrected according to the regression quantification coefficient to obtain the corrected control index. The corrected control index and the long-term control index are linked and integrated to generate control linkage parameters. The step of obtaining the regression quantification coefficient by performing regression gradient detection according to the long-term regulation index includes: performing time-series scanning of the long-term regulation index to identify stagnant segments of schemes that have not triggered regulation to form a stagnant segment sequence; spontaneously identifying the stagnant segment sequence according to stagnant trigger characteristics to generate spontaneous stagnant markers; classifying and weighting the stagnant segment sequence according to the spontaneous stagnant markers to generate a weighted stagnant distribution; and determining the regression quantification coefficient based on the cumulative intensity of the weighted stagnant distribution. By integrating the aforementioned regulation and linkage parameters with the fatigue load assessment spectrum, the optimal activation time of the optimal solution is identified through benefit stage comparison. Based on the activation time of the optimal solution, the optimal rehabilitation training program is selected and rehabilitation training regulation instructions are output.

2. The method according to claim 1, characterized in that, The step of using the recovery ability assessment scale to locate and train weak functional areas includes: The scores of each item in the recovery capacity assessment scale are extracted according to the functional dimensions to generate a functional score distribution table; The function score distribution table is subjected to continuous decline detection during multiple evaluations to generate continuous decline dimension labels. The inflection point density distribution is generated by cross-dimensional density superposition using the continuously descending dimension markers. Weak functional areas in training are identified by locating functional dimensions with scores below a threshold based on the inflection point density distribution.

3. The method according to claim 1, characterized in that, The step of generating a fatigue load assessment spectrum by performing bilateral fatigue asymmetric quantification on the weak functional areas of the training includes: Extract bilateral limb movement amplitude time series from the aforementioned weak functional areas to generate bilateral amplitude time series groups; The fatigue persistence feature is extracted from the bilateral amplitude time series to generate a fatigue persistence distribution. Asymmetric fatigue coefficients are generated by performing left-right asymmetric joint weighting based on the bilateral amplitude time series and the fatigue persistence distribution. A fatigue load assessment spectrum is constructed based on the asymmetric fatigue coefficient.

4. The method according to claim 1, characterized in that, The step of performing diminishing returns analysis on the fatigue load assessment spectrum and the scheme adaptation margin to generate diminishing returns markers includes: The fatigue load assessment spectrum and the scheme adaptation margin are numerically compared on a training cycle basis to generate a cycle comparison sequence. The training phase, which identifies the synchronous deterioration of fatigue load and scheme adaptation margin in the cycle comparison sequence, generates synchronous deterioration intervals. The synchronous deterioration intervals are assigned weighted decreasing coefficients according to their coverage and intensity to generate classification decreasing label groups; Based on the coverage and intensity weighting of the aforementioned classification of decreasing marker groups, diminishing benefit markers are generated.

5. The method according to claim 1, characterized in that, The step of applying a threshold correction to the real-time control indicator according to the regression quantification coefficient to obtain the corrected control indicator includes: The regression quantization coefficients are segmented into linearized segments to generate mild regression segments and abrupt regression segments; The mild regression segment and the abrupt regression segment are subjected to a graded step size correction allocation to generate a segmented correction sequence; The modified threshold is generated by superimposing the segmented modified sequence onto the original threshold of the real-time control indicator; The modified control index is obtained by updating the real-time control index using the modified threshold to apply over-limit constraints.

6. The method according to claim 1, characterized in that, The process of integrating the regulation and linkage parameters with the fatigue load assessment spectrum to identify the optimal solution activation timing through benefit stage comparison includes: The aforementioned control linkage parameters are mapped to the fatigue load assessment spectrum to generate a dynamic evaluation table of training benefits. The training benefit dynamic evaluation table is used to locate cross-scheme benefit cross-intervals and generate cross-interval groups. The physical endurance threshold for the corresponding time period in the fatigue load assessment spectrum is extracted using the cross-interval group, and the time period of insufficient physical strength is screened out to generate a valid activation candidate table; The optimal activation time is determined based on the effective activation candidate table.

7. The method according to claim 2, characterized in that, The step of generating continuous decline dimension labels by performing continuous decline detection on the functional score distribution table through multiple evaluations includes: For each functional dimension in the functional score distribution table, scores are extracted segment by segment according to the evaluation period to generate a periodic score sequence; The periodic score sequence is used to identify segments where the cross-dimensional magnitude difference continues to widen, forming a difference widening region marker. The inflection point risk classification and annotation are generated using the aforementioned difference expansion area marker; Generate a continuously declining dimension marker based on the inflection point risk annotation.

8. The method according to claim 1, characterized in that, The step of classifying and weighting the stagnation segment sequence according to the spontaneous stagnation marker to generate a weighted stagnation distribution includes: The stagnation sequence is segmented and fitted according to the duration of stagnation to generate a baseline stagnation curve; The residual stagnation sequence is generated by performing difference separation between the stagnation segment sequence and the benchmark stagnation curve. The passive stagnation sequence is passively oriented according to the spontaneous stagnation marker to generate a set of passive stagnation segments. The passive stagnation segment set is assigned a high weight to generate a weighted stagnation distribution.

9. An intelligent control system for postoperative rehabilitation training programs after cerebral hemorrhage, characterized in that, include: The scale construction unit is used to acquire limb movement data and neurological function scores and perform functional correlation analysis to obtain a recovery ability assessment scale. Based on the recovery ability assessment scale, the limb movement data and neurological function scores are verified for completeness of adaptation to obtain a solution adaptation margin. The assessment and analysis unit is used to locate the weak functional areas of training using the recovery ability assessment scale, and to generate a fatigue load assessment spectrum by performing bilateral fatigue asymmetry quantification on the weak functional areas of training. The index construction unit is used to perform diminishing returns analysis on the fatigue load assessment spectrum and the scheme adaptation margin to generate diminishing returns markers, identify continuous stagnation intervals according to the diminishing returns markers to generate the urgency of scheme adjustment, and construct a gradient control index table according to the urgency of scheme adjustment. The parameter integration unit is used to split the gradient control index table according to the response time to generate immediate control index and long-term control index, perform regression gradient detection according to the long-term control index to obtain regression quantification coefficient, perform threshold correction on the immediate control index according to the regression quantification coefficient to obtain the corrected control index, and combine the corrected control index and the long-term control index to generate control linkage parameters. The step of obtaining the regression quantification coefficient by performing regression gradient detection according to the long-term regulation index includes: performing time-series scanning of the long-term regulation index to identify stagnant segments of schemes that have not triggered regulation to form a stagnant segment sequence; spontaneously identifying the stagnant segment sequence according to stagnant trigger characteristics to generate spontaneous stagnant markers; classifying and weighting the stagnant segment sequence according to the spontaneous stagnant markers to generate a weighted stagnant distribution; and determining the regression quantification coefficient based on the cumulative intensity of the weighted stagnant distribution. The scheme output unit is used to integrate the control linkage parameters with the fatigue load assessment spectrum to compare the benefits of different stages, identify the optimal scheme activation time, and select the optimal rehabilitation training scheme according to the optimal scheme activation time, and output rehabilitation training control instructions.

Images