CN-122024999-A - Personalized rehabilitation scheme optimization method based on bone healing technology

CN122024999ACN 122024999 ACN122024999 ACN 122024999ACN-122024999-A

Abstract

The invention provides a personalized rehabilitation scheme optimization method and a personalized rehabilitation scheme optimization system based on a bone healing technology, and relates to the technical field of medical rehabilitation, wherein the method comprises the steps of obtaining CT images of a bone region to be detected, and outputting classification probability of early, medium and mature healing of the bone region to be detected through a three-dimensional residual neural network; the method comprises the steps of calculating the non-purity of the keni and the trust coefficient of a healing stage based on each classification probability, determining the fine healing stage, primarily selecting actions adapting to the fine healing stage from a rehabilitation action library to form a candidate rehabilitation action set, combining the candidate rehabilitation actions in the candidate rehabilitation action set by combining individual preferences of a patient to form a primary personalized rehabilitation scheme, obtaining physiological data and motion sequence data of the patient in the process of executing the primary personalized rehabilitation scheme, calculating a motion quality score reflecting the motion efficiency and the physiological load of the patient, and optimizing the scheme based on the motion quality score to obtain the optimal personalized rehabilitation scheme.

Inventors

XU CHENG
XIONG DOU
ZHOU HAO
WANG KAIXUAN
CHEN BOWEI
ZHANG HAO
TANG PEIFU
LI JIANTAO
ZHANG WEI
JIA ZHENGFENG
LI MENG
ZHANG ZICHENG
ZHAO ZIXIN
MO FUHAO
XIA YANWEI

Assignees

中国人民解放军总医院第四医学中心

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (9)

1. A personalized rehabilitation regimen optimization method based on bone healing technology, comprising: s1, acquiring CT images of a bone region to be detected; S2, based on the CT image, outputting classification probability that the bone region to be detected belongs to early healing, mid-term healing and mature healing through a three-dimensional residual neural network; S3, determining a fine healing stage to which the bone region to be detected belongs by calculating the non-purity of the keni and the trust coefficient of the healing stage based on the classification probabilities; S4, preliminarily selecting a plurality of candidate rehabilitation actions adapting to the fine healing stage from a preset rehabilitation action library to form a candidate rehabilitation action set; S5, combining all candidate rehabilitation actions in the candidate rehabilitation action set according to individual preferences of the patient to form a preliminary personalized rehabilitation scheme; S6, acquiring physiological data and motion sequence data of a patient in the process of executing the primary personalized rehabilitation scheme, and calculating a motion quality score reflecting the motion efficiency of the patient and the physiological load of the patient based on the physiological data and the motion sequence data; And S7, based on the motion quality scores, carrying out optimization adjustment on the preliminary personalized rehabilitation scheme to obtain an optimal personalized rehabilitation scheme.
2. The personalized rehabilitation regimen preference method based on bone healing technology according to claim 1, characterized in that S2 comprises in particular: s201, cutting the CT image in a voxel cutting mode to obtain a three-dimensional input image conforming to the input size of a three-dimensional residual neutral network; S202, inputting the three-dimensional input image into a feature extraction network comprising a plurality of three-dimensional convolution layers and residual connection, extracting local space texture information by utilizing a three-dimensional convolution kernel, realizing feature accumulation and propagation through a residual structure, and extracting a primary bone healing feature map; S203, enhancing the primary bone healing feature map by using a channel attention mechanism to obtain a channel enhanced bone healing feature map; s204, performing enhancement treatment on the channel enhanced bone healing feature map by using a spatial attention mechanism to obtain a combined enhanced bone healing feature map; s205, based on the combined enhanced bone healing feature map, combining attention enhancement features with different depths, and utilizing a self-adaptive feature fusion module to realize weighted fusion of coarse-scale and fine-scale features to obtain a three-dimensional fusion feature map; s206, outputting the classification probability of the bone region to be detected belonging to early healing, mid-term healing and mature healing through a softmax classification layer based on the three-dimensional fusion feature map.
3. The personalized rehabilitation regimen preference method based on bone healing techniques according to claim 2, characterized in that S204 comprises in particular: s2041, carrying out average pooling and maximum pooling on the channel-enhanced bone healing feature map in the channel dimension to obtain an average pooling map and a maximum pooling map; S2042, performing splicing processing on the average pooling graph and the maximum pooling graph, and generating a spatial attention weight through convolution operation; And S2043, carrying out enhancement treatment on the channel enhanced bone healing characteristic map based on the spatial attention weight to obtain the combined enhanced bone healing characteristic map.
4. The personalized rehabilitation regimen preference method based on bone healing technology according to claim 1, characterized in that S3 comprises in particular: S301, calculating the Indonesia based on early healing probability, mid healing probability and mature healing probability; s302, judging whether the Arrhenius purity is less than a preset Arrhenius purity or not, if so, entering S303, otherwise, entering a manual confirmation mode; s303, determining a maximum probability value and a target probability value of a healing stage before the maximum probability value from the early healing probability, the middle healing probability and the mature healing probability; s304, calculating a healing stage trust coefficient of a healing stage corresponding to the maximum probability value based on the maximum probability value and the target probability value; S305, judging whether the healing stage trust coefficient is larger than a preset healing stage trust coefficient, if so, taking the healing stage corresponding to the maximum probability value as the final fine healing stage of the bone region to be detected, and if not, taking the healing stage corresponding to the target probability value as the final fine healing stage of the bone region to be detected.
5. The personalized rehabilitation regimen preference method based on bone healing technology according to claim 1, characterized in that S4 comprises in particular: S401, selecting a plurality of candidate rehabilitation actions adapting to the fine healing stage from the rehabilitation action library; S402, determining a safety boundary range of the fine healing stage; and S403, eliminating the candidate rehabilitation actions which do not meet the safety boundary range from the candidate rehabilitation actions to form the candidate rehabilitation action set.
6. The personalized rehabilitation regimen preference method based on bone healing technology according to claim 1, characterized in that S5 comprises in particular: S501, according to the individual preference of the patient, eliminating candidate rehabilitation actions conflicting with the preference of the patient from the candidate rehabilitation action set to form a target candidate rehabilitation action set; s502, calculating the safety scores of all candidate rehabilitation actions in the target candidate rehabilitation action set, and sequencing the candidate rehabilitation actions corresponding to the safety scores according to the sequence from high to low to obtain a candidate rehabilitation action sequencing list; S503, according to training logic, rehabilitation actions are sequentially selected from the candidate rehabilitation action sequencing list, and the preliminary personalized rehabilitation scheme is formed.
7. The personalized rehabilitation regimen preference method based on bone healing technology according to claim 1, characterized in that S6 comprises in particular: s601, acquiring physiological data of the patient including heart rate and electromyographic signals through wearable equipment; s602, acquiring a video stream of the patient for executing standard rehabilitation actions through a depth sensor, and extracting motion sequence data of a bone region to be detected in the video stream by using a YOLOv-ShuffleNetV model; S603, performing time stamp alignment processing on the physiological data and the motion sequence data; s604, inputting the physiological data and the motion sequence data which are subjected to alignment treatment into a two-way long-short-term memory network model, and analyzing a time sequence relation between the motion sequence and physiological response to obtain a multidimensional feature vector for representing the physiological adaptability and the motion state matching degree of a patient; s605, extracting motion efficiency characteristic parameters and physiological load characteristic parameters of the multidimensional characteristic vector; S606, calculating a motion quality score reflecting the motion efficiency and the physiological load of the patient through a scoring function according to the motion efficiency characteristic parameter and the physiological load characteristic parameter.
8. The personalized rehabilitation regimen preference method based on bone healing technology according to claim 1, characterized in that S7 comprises in particular: s701, judging whether the motion quality score is larger than a preset motion quality score, if so, judging that the current primary personalized rehabilitation scheme is suitable for the patient, entering S702, otherwise, judging that the current primary personalized rehabilitation scheme is not suitable for the patient, entering S703; S702, executing an excitation feedback mode, increasing the action intensity in the primary personalized rehabilitation scheme and prolonging the training time or the training group number of the action under the condition that the safety boundary condition is met, and entering S704; s703, executing an intervention reminding mode, reducing the load parameters of actions in the primary personalized rehabilitation scheme and reducing the repeated times or the training group number of the actions, and entering S704; And S704, dynamically updating the action parameters of the primary personalized rehabilitation scheme based on the excitation feedback mode result or the intervention reminding mode result to obtain the optimal personalized rehabilitation scheme.
9. The personalized rehabilitation regimen-preferred method based on bone healing technology according to claim 8, further comprising, after said S7: In the case that the sports quality scores of 3 continuous times are all larger than the preset sports quality score, judging that the current healing stage is possibly not matched with the current healing state any more, and consulting whether a doctor can enter the training of the next healing stage or not; and under the condition that the motion quality scores of 3 continuous times are less than or equal to the preset motion quality score, judging a high risk scheme of the current scheme, reducing the current healing stage to the previous healing stage or reselecting the rehabilitation action with the minimum load in the rehabilitation action library.

Description

Personalized rehabilitation scheme optimization method based on bone healing technology Technical Field The invention relates to the technical field of medical rehabilitation, in particular to a personalized rehabilitation scheme optimization method based on a bone healing technology. Background The rehabilitation process after fracture has obvious stage, individual difference and long-term property, and the function recovery effect is not only influenced by the healing speed of bone tissues, but also is closely related to muscle strength increase, physiological load tolerance and training action quality. Along with the development of wearable sensors, medical image reconstruction and deep learning evaluation technologies, rehabilitation scheme optimization based on intelligent prediction gradually becomes a research focus in the fields of rehabilitation engineering and sports medicine. The existing research generally divides the fracture healing condition into stages based on medical images, bone density parameters or traditional clinical evaluation scales (such as VAS, harris score and the like), and utilizes myoelectric signals, an inertial measurement unit or gait analysis and other means to judge the quality of rehabilitation actions. In the aspect of intelligent prediction, a convolutional neural network is adopted in some methods for identifying fracture healing stages, or LSTM, random forest and other machine learning models are used for predicting rehabilitation progress, and in the aspect of scheme recommendation, an existing system tries to match rehabilitation projects according to exercise task difficulty, physiological energy consumption or repeated training effects. However, the existing method mainly relies on static or local features in the bone healing image to judge, lacks deep feature extraction of evolution rules of the bone healing stage, and is difficult to accurately give a prediction result of the current healing state of a patient. Meanwhile, the action quality, fatigue level and training response of a patient in the rehabilitation process have the characteristic of continuous dynamic change, but the existing method generally lacks a real-time feedback and self-adaptive adjustment mechanism, and the training content cannot be flexibly optimized according to the staged evaluation result, so that the rehabilitation scheme is difficult to realize closed-loop management and personalized regulation and control. Disclosure of Invention In order to solve the problems that the existing method mainly depends on static or local characteristics in a bone healing image to judge, deep characteristic extraction of evolution rules of a bone healing stage is lacking, and a prediction result of the current healing state of a patient is difficult to accurately give. Meanwhile, the action quality, fatigue level and training response of a patient in the rehabilitation process have the characteristic of continuous dynamic change, but the existing method generally lacks a real-time feedback and self-adaptive adjustment mechanism, and training content cannot be flexibly optimized according to a staged evaluation result, so that the technical problem that a rehabilitation scheme is difficult to realize closed-loop management and personalized regulation is solved. The technical scheme provided by the embodiment of the invention is as follows: first aspect: the embodiment of the invention provides a personalized rehabilitation scheme optimization method based on a bone healing technology, which comprises the following steps: s1, acquiring CT images of a bone region to be detected; S2, based on the CT image, outputting classification probability that the bone region to be detected belongs to early healing, mid-term healing and mature healing through a three-dimensional residual neural network; S3, determining a fine healing stage to which the bone region to be detected belongs by calculating the non-purity of the keni and the trust coefficient of the healing stage based on the classification probabilities; S4, preliminarily selecting a plurality of candidate rehabilitation actions adapting to the fine healing stage from a preset rehabilitation action library to form a candidate rehabilitation action set; S5, combining all candidate rehabilitation actions in the candidate rehabilitation action set according to individual preferences of the patient to form a preliminary personalized rehabilitation scheme; S6, acquiring physiological data and motion sequence data of a patient in the process of executing the primary personalized rehabilitation scheme, and calculating a motion quality score reflecting the motion efficiency of the patient and the physiological load of the patient based on the physiological data and the motion sequence data; And S7, based on the motion quality scores, carrying out optimization adjustment on the preliminary personalized rehabilitation scheme to obtain an optimal personalized rehabil