An evaluation method and device for postoperative home rehabilitation training of lung cancer

Abstract

The application provides an evaluation method and device for postoperative home rehabilitation training of lung cancer, effectively solving the problem that the prior art lacks accurate and effective technical solutions for postoperative home rehabilitation training of lung cancer. The method comprises the following steps: acquiring, based on a wearable device and a mobile phone camera, multi-source heterogeneous data of a user after lung cancer surgery in a home rehabilitation training stage within a preset evaluation window, and extracting multi-dimensional features in the multi-source heterogeneous data; calculating a plurality of function domain scores of a plurality of function domains pre-divided based on the multi-dimensional features, and cooperatively calculating multi-dimensional parameter values of each candidate rehabilitation training project of the next stage of the user based on the multi-dimensional features and the plurality of function domain scores; based on the multi-dimensional parameter values, calculating an adaptation degree prediction value of each candidate rehabilitation training project in the next stage to obtain a next stage training project and a corresponding training dose parameter of the user in the home rehabilitation training stage.

Description

Technical Field

[0001] This application relates to the field of rehabilitation training technology, and more specifically, to an assessment method and device for home rehabilitation training after lung cancer surgery. Background Technology

[0002] Postoperative home rehabilitation for lung cancer requires various training exercises, including breathing, chest wall movement, upper limb shoulder and chest range of motion, and walking endurance. However, inappropriate intensity, unsafe environment, or poor execution can easily lead to fatigue, hypoxia, falls, and decreased compliance. Existing home rehabilitation systems generally suffer from three prominent problems: training plans are issued based on postoperative time or fixed templates, lacking quantitative assessment of the patient's actual "here and now" condition and home training conditions; reliance on a single data source makes it difficult to simultaneously consider breathing, movement, endurance, and safety; and non-contact assessments are limited to monitoring, failing to integrate monitoring results, training performance, and home environment resources / risks to form an actionable prediction of the next stage of training content.

[0003] Technicians have proposed three relatively strong technical approaches: robust respiratory feature extraction based on optical flow and spectral analysis of multiple anatomical regions (forehead, cheeks, upper chest, shoulders, etc.) from ordinary RGB video, combined with one-dimensional convolutional temporal models and SNR quality weighting; real-time identification of respiratory phase based on nasal alar / nostril ROI using the coldest pixel temperature sequence, MAD adaptive threshold, and hysteresis state machine; and estimation of respiratory flow based on joint modeling of a single differential pressure sensor and battery voltage, using integral trends to identify improper wearing or leakage. However, these technologies remain at the level of single-point applications such as "respiratory monitoring," "thermal imaging detection," or "leakage identification," and have not yet formed a unified technical framework for post-lung cancer surgery home rehabilitation.

[0004] Therefore, it is necessary to propose a new solution that integrates wearable devices, multimodal machine vision (integrating RGB, thermal imaging, etc.), training process data, and objective identification of the home environment into a closed loop. This solution can identify the patient's current physiological and motor abilities, determine the home safety training conditions, and, in conjunction with past training performance, predict the training content, priority, and dosage for the next stage, thus achieving a leap from passive monitoring to adaptive training recommendations. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide an assessment method and device for home rehabilitation training after lung cancer surgery, which effectively solves the problem that the existing technology lacks an accurate and effective technical solution for home rehabilitation training after lung cancer surgery.

[0006] In a first aspect, embodiments of this application provide an assessment method for home rehabilitation training after lung cancer surgery, the method comprising: Based on wearable devices and mobile phone cameras, multi-source heterogeneous data of lung cancer postoperative users in the home rehabilitation training stage are acquired within a preset assessment window, and multi-dimensional features are extracted from the multi-source heterogeneous data. Based on the multi-dimensional features, multiple functional domain scores of the pre-divided functional domains are calculated, and based on the multi-dimensional features and multiple functional domain scores, multi-dimensional parameter values ​​of each candidate rehabilitation training item for the user's next stage are calculated collaboratively. Based on the multi-dimensional parameter values, the predicted fit of each candidate rehabilitation training item in the next stage is calculated to obtain the next stage training items and corresponding training dosage parameters for the user in the home rehabilitation training stage.

[0007] In conjunction with the first aspect, this application provides a first possible implementation of the first aspect, wherein extracting multi-dimensional features from the multi-source heterogeneous data includes: Pre-divide standard multi-source heterogeneous data into multiple standard data types; different standard data types correspond to different feature extraction methods; Identify the standard data type corresponding to the user's multi-source heterogeneous data, and perform the corresponding feature extraction method to obtain the multi-dimensional features.

[0008] In conjunction with the first aspect, this application provides a second possible implementation of the first aspect, wherein the multi-dimensional features include at least respiratory features and motion features; The multi-dimensional features are obtained by performing the corresponding feature extraction method, including: Calculate the dense optical flow displacement amplitude or inter-frame displacement sequence between adjacent frames of each region of interest in the video data of the multi-source heterogeneous data. The displacement amplitude or inter-frame displacement sequence is processed to obtain an estimated visible light respiration rate, which is then used as a respiration feature. By combining the temporal sequence of key points on the human body, at least one preset multiple action sub-features are extracted as action features.

[0009] In conjunction with the first aspect, this application provides a third possible implementation of the first aspect, wherein the multi-dimensional features include at least thermal imaging respiratory features; The multi-dimensional features are obtained by performing the corresponding feature extraction method, including: Target detection is performed on the nasal wing and / or nostril region of the thermal imaging data in the multi-source heterogeneous data, and region tracking is performed between adjacent detection frames; The temperature sequence of the coldest pixel in the tracked area is extracted, and the thermal imaging respiration rate estimate is extracted from the temperature sequence of the coldest pixel and used as the thermal imaging respiration feature.

[0010] In conjunction with the first aspect, this application provides a fourth possible implementation of the first aspect, wherein the multi-dimensional parameter values ​​of each candidate rehabilitation training item for the user's next stage are calculated collaboratively based on the multi-dimensional features and multiple functional domain scores, including: Multi-dimensional standard parameters are set based on candidate rehabilitation training programs, user's physical condition, current family environment, actual skill mastery, and adaptation status. The multi-dimensional parameter values ​​are obtained by performing calculations on the multi-dimensional features and the scores of multiple functional domains according to the corresponding multi-dimensional standard parameters.

[0011] In conjunction with the first aspect, this application provides a fifth possible implementation of the first aspect, wherein obtaining the user's next-stage training project and corresponding training dose parameters includes: Candidate rehabilitation training programs are ranked based on their fit prediction values ​​to select the user's next stage of training. Based on the user's current training performance, current symptoms, and current home environment, the training dosage parameters corresponding to the next stage of training are dynamically determined.

[0012] In conjunction with the first aspect, this application provides a sixth possible implementation of the first aspect, wherein calculating multiple functional domain scores for pre-divided multiple functional domains based on the multi-dimensional features includes: For multiple functional domains, corresponding features are extracted from the multi-dimensional features, and the features are normalized. The normalized features are processed by a pre-set functional domain calculation model to obtain the corresponding functional domain score.

[0013] Secondly, embodiments of this application provide an assessment device for home rehabilitation training after lung cancer surgery, the device comprising: The acquisition module is used to acquire multi-source heterogeneous data of lung cancer postoperative users in the home rehabilitation training stage within a preset assessment window based on wearable devices and mobile phone cameras, and extract multi-dimensional features from the multi-source heterogeneous data. The scoring module is used to calculate multiple functional domain scores for multiple pre-divided functional domains based on the multi-dimensional features, and to collaboratively calculate the multi-dimensional parameter values ​​of each candidate rehabilitation training item for the user's next stage based on the multi-dimensional features and the multiple functional domain scores. The calculation module is used to calculate the predicted fit value of each candidate rehabilitation training item in the next stage based on the multi-dimensional parameter values, so as to obtain the next stage training items and corresponding training dosage parameters for the user in the home rehabilitation training stage.

[0014] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of an assessment method for home rehabilitation training after lung cancer surgery as described in any one of the claims.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of an assessment method for postoperative home rehabilitation training for lung cancer as described in any one of the claims.

[0016] This application provides an assessment method for home rehabilitation training after lung cancer surgery. The method first acquires multi-source heterogeneous data of a user undergoing home rehabilitation training within a preset assessment window using a wearable device and a mobile phone camera, and extracts multi-dimensional features from the multi-source heterogeneous data. Then, based on the multi-dimensional features, it calculates multiple functional domain scores for pre-divided functional domains, and collaboratively calculates multi-dimensional parameter values ​​for each candidate rehabilitation training item in the next stage based on the multi-dimensional features and the multiple functional domain scores. Finally, based on the multi-dimensional parameter values, it calculates the predicted fit value for each candidate rehabilitation training item in the next stage, thereby obtaining the next stage training items and corresponding training dosage parameters for the user in the home rehabilitation training stage. Based on the above methods, this application can obtain not only contact data such as heart rate, blood oxygen, and activity level, but also non-contact objective information such as respiratory rhythm, chest rise and fall, movement quality, balance risk, and current home environment. This solves the problem of incomplete patient condition assessment in existing home rehabilitation systems. It also addresses the inability of existing programs to objectively determine whether a patient can do the exercises at home, whether they are suitable, and which type is safer. Training recommendations are no longer solely dependent on post-operative time or subjective questionnaires, but are constrained by environmental factors such as trainable space, obstacle density, seating conditions, lighting conditions, handrail / support conditions, and ground risks. Furthermore, it addresses the inability to connect "how well I've been training recently" with "what should I train next," enabling the system to predict more suitable training content, priority, and dosage progression for the next stage based on information such as previous training completion rate, movement quality, symptom burden, fatigue interruption, and improvement trends. Finally, it addresses the insufficient robustness of single visual modalities in low light, skin color differences, slight head movements, or occlusion conditions by utilizing visible light multi-ROI. Quality-weighted fusion of optical flow, thermal imaging nasal alar temperature signals, and optional differential pressure flow signals improves the stability of postoperative home rehabilitation assessment. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an assessment method for postoperative home rehabilitation training of lung cancer provided in an embodiment of this application is shown. Figure 2 A schematic diagram of the process for extracting multi-dimensional features provided in an embodiment of this application is shown; Figure 3 This illustration shows a flowchart of obtaining the user's next-stage training project and corresponding training dose parameters according to an embodiment of this application. Figure 4 This paper shows a structural block diagram of an assessment device for home rehabilitation training after lung cancer surgery, provided in an embodiment of this application. Figure 5 A structural block diagram of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0020] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0022] Existing technical solutions have rigid training plans and limited data. RGB, thermal imaging, and differential pressure respiratory analysis are only used in single points, which makes postoperative home training for lung cancer patients prone to risks due to inappropriate intensity or environment, and thus cannot achieve the expected results.

[0023] Based on this, this application provides an assessment method and apparatus for home rehabilitation training after lung cancer surgery, which will be described below through embodiments.

[0024] Example 1 To facilitate understanding of this embodiment, a detailed description of an assessment method for postoperative home rehabilitation training of lung cancer patients, as disclosed in this application, will be provided first. For example... Figure 1 The diagram illustrates a flowchart of an assessment method for home rehabilitation training after lung cancer surgery. This application provides an assessment method for home rehabilitation training after lung cancer surgery, the method comprising: S101. Based on wearable devices and mobile phone cameras, acquire multi-source heterogeneous data of lung cancer postoperative users in the home rehabilitation training stage within a preset assessment window, and extract multi-dimensional features from the multi-source heterogeneous data. S102. Calculate multiple functional domain scores for the pre-divided multiple functional domains based on the multi-dimensional features, and collaboratively calculate the multi-dimensional parameter values ​​of each candidate rehabilitation training item for the user's next stage based on the multi-dimensional features and the multiple functional domain scores. S103. Based on the multi-dimensional parameter values, calculate the predicted fit of each candidate rehabilitation training item in the next stage, so as to obtain the next stage training items and corresponding training dosage parameters for the user in the home rehabilitation training stage.

[0025] In step S101, this application acquires multi-source heterogeneous data from a user undergoing home rehabilitation after lung cancer surgery within a preset assessment window using a wearable device such as a smartwatch, a clip-on or wrist-mounted pulse oximeter, and a mobile phone camera. The preset assessment window can be set according to the user's actual needs, specifically representing a time period for assessment. The multi-source heterogeneous data also includes physiological data collected by the wearable device, visible light patient video data used to capture patient assessment actions, home scene images or videos used to capture home training scenarios, and historical training performance data recorded by the intelligent rehabilitation training system. When a thermal imaging device or a breathing training accessory with a flow channel is present, thermal imaging data and / or differential pressure breathing training accessory data are also acquired. The physiological data... This includes data collected by wearable devices such as heart rate, resting heart rate, heart rate recovery, blood oxygen saturation, physical activity intensity, steps, posture, sleep, upper limb activity, trunk activity, training duration, and number of training interruptions; visible light patient video data includes training video data of the patient's face, upper chest, shoulders, and whole body, as well as home scene video data. It can also include mobile phone-guided patient training movements such as: sitting, standing, stepping in place, sitting-standing movements, and scanning data of walking paths, training areas, and chairs; thermal imaging data is obtained by acquiring thermal images of the patient's nasal alae / nostrils, used to extract respiratory phase and rhythm under low light conditions, privacy-sensitive environments, or poor visible light quality. Specifically, it employs small nasal alae detection and Kalman tracking, making it particularly suitable for embedded edge devices; differential pressure respiratory training accessory data is collected when the user uses a respiratory muscle trainer, respiratory training mask, or training accessory with a flow channel, allowing the system to collect differential pressure ΔP and battery voltage V. b Alternatively, the power supply status is used to estimate inspiratory and expiratory flow rates, training effectiveness, and accessory wearing / leakage status during training. The specific technical approach is consistent with that of a single differential pressure sensor for estimating respiratory flow and leakage. Historical training performance data is extracted from training logs, including completion rate, movement attainment rate, number of repetitions, number of sets, time to target achievement, reason for training termination, subjective fatigue, dyspnea score, pain score, and recovery time for each training item. After obtaining the above multi-source heterogeneous data, the multi-dimensional features are extracted after preprocessing. In the specific implementation of step S101, one embodiment is as follows: Figure 2 As shown, multi-dimensional features are extracted from the multi-source heterogeneous data, including: S1011. Pre-divide standard multi-source heterogeneous data into multiple standard data types; different standard data types correspond to different feature extraction methods; S1012. Identify the standard data type corresponding to the user's multi-source heterogeneous data, and perform the corresponding feature extraction method to obtain the multi-dimensional features.

[0026] In steps S1011-S1012, this application pre-divides standard multi-source heterogeneous data into multiple standard data types. Standard data types are defined based on data names, and corresponding feature extraction methods are pre-configured for each standard data type. Different standard data types correspond to different feature extraction methods. After obtaining the user's multi-source heterogeneous data, the standard data type corresponding to the user's multi-source heterogeneous data is identified, and the corresponding feature extraction method is called to execute the corresponding feature extraction method to obtain the multi-dimensional features. Before feature extraction, the multi-source heterogeneous data is preprocessed to improve the effectiveness and accuracy of the extracted multi-dimensional features. The multi-dimensional features include physiological state features, action features, and respiratory state features. The physiological state features are extracted based on physiological data. If thermal imaging data exists, thermal imaging respiratory features are also extracted. If differential pressure breathing training accessory data exists, respiratory flow features and leakage state features are also extracted. Target recognition, region segmentation, and / or geometric measurement are performed on the home scene image or video data to identify scene objects and their attributes related to the training implementation conditions, and candidate training items are selected. T j Extracting environmental resource characteristics e j,k and environmental risk characteristics h j,l This allows for the assessment not only of "whether a user can practice," but also "whether this practice method is suitable for the user's home environment." Therefore, this application can transform home scene recognition results into training programs tailored to specific needs. T j The computable environmental representation, together with the functional status characteristics of the patient obtained through wearable devices and machine vision assessment, and the historical training performance in the intelligent rehabilitation training system, are used to predict the training content for the next stage.

[0027] The respiratory state features in this application involve evaluating the quality of the multi-source heterogeneous data sources, and fusing at least the respiratory features based on the quality evaluation results of each data source. Specifically, based on at least one of the signal-to-noise ratio, target detection confidence, region tracking stability, motion blur degree, and leakage state features, quality weights are assigned to visible light respiratory features, thermal imaging respiratory features, and differential pressure respiratory training accessory features. q vis , q th , q prsThe data is then weighted and fused according to the aforementioned quality weights to obtain a fused respiratory rate estimate, which is used as the user's respiratory state feature RR. ; Among them, RR vis RR th and RR prs These are the estimated respiratory rates for visible light, thermal imaging, and differential pressure accessory modes, respectively. q vis , q th and q prs These represent the quality weights for the corresponding modalities; ε is a small constant to prevent the denominator from being zero. When a modality is not enabled, the corresponding weight is recorded as 0. In addition to respiratory rate, this application also extracts the following rehabilitation assessment features: surrogate value of chest rise and fall amplitude, inspiratory-to-expiratory ratio, rhythm stability, number of cough events, rhythm of marching in place, number of sit-to-stand repetitions, time to complete a single sit-to-stand repetition, surrogate value of walking speed, range of motion of the upper limbs and shoulders, trunk swing amplitude, heart rate recovery after training, minimum blood oxygenation during training, and training interruption rate.

[0028] The breathing flow characteristics and leakage state characteristics mentioned above are obtained based on the following feature extraction method: the user wears or connects a training accessory with a flow channel during breathing training, and this application utilizes differential pressure. ΔP ( t ) and battery voltage V b ( t Estimate the flow state, and let the training attachment inflow estimation function be... f 1(·), the outflow estimation function is f 2(·), then estimate the respiratory flow rate. Q resp,e ( t ) is defined as: ; ; ; The above figures represent inspiratory flow rate, expiratory flow rate, and respiratory flow rate, respectively. A judgment window is calculated to determine if the accessory leaks, is improperly worn, or if training cooperation is insufficient. t 1, t Integral quantity within 2] I leak : ; This involves calculating the differential pressure signal, inspiratory flow rate, and / or expiratory flow rate within a preset judgment window to obtain the flow integral value, flow offset, or flow balance relationship within the preset judgment window.I leak If the value is less than the set threshold, it is determined that there may be problems such as leakage of training accessories, poor sealing of the mask, or improper training actions by the user.

[0029] The environmental resource characteristics mentioned therein e j,k At least one or more of the following are included: available training clearance area, walkable path length, seat fit, availability of armrests or support structures, step availability, and lighting conditions, wherein the seat fit is determined at least by seat height, backrest presence, and armrest presence; the environmental risk characteristics are also included. h j,l It includes at least one or more of the following: obstacle density, risk of slippery ground, risk of occlusion, risk of insufficient lighting, and risk of insufficient camera field of view coverage; specifically, for candidate training items. T j Training space metrics are available. e space,j Defined as: ; in, A usable,j For training projects T j The identified area of ​​passable, standing, and trainable clearance zones; A ref,j To complete the training program T j The required reference area. e space,j For environmental resource characteristics e j,k A specific component; while the density of obstacles in the home environment. h obs,j Defined as: ; In the formula, h obs,j For training projects T j The density of obstacles within the corresponding training area; N obs,j The number of obstacles within the training area; A scene,j If the training region is the scene area, then the h obs,j Environmental risk characteristics h j,l A specific component; seat fit e chair,j Defined as: ; In the formula, e chair,j For training projects T j Seat fit; I arm,j Indicate whether the candidate seats have armrests; I back,j Indicate whether the candidate seat has a backrest; H seat,j Indicates the seat height of the candidate seats; H 0,j Indicates training project T j Target adaptation height; σ H If it is a height tolerance parameter, then e chair,j For environmental resource characteristics e j,k A specific component. For each candidate training item T j Define environment adaptation score E j : ; In the formula, E j For training projects T j Environmental adaptation score; e j,k For training projects T j The kth environmental resource characteristic includes e space,j , e chair,j One or more of the following: handrail availability, step availability, and lighting suitability; h j,l For training projects T j The first environmental risk characteristic includes h obs,j One or more of the following: risk of slippery ground, risk of obstruction, and risk of insufficient sunlight; a j,k This represents the positive weight of environmental resource characteristics on training items; b j,l Indicates the environmental risk characteristics for training projects T j negative weights, E j The larger the value, the more suitable the current home environment is for carrying out the training program. Tj For example, indoor walking training relies more on... e space,j and h obs,j Sitting and standing training relies more on e chair,j In addition to the anti-slip conditions of the ground; step training relies more on the usability of the steps and the stability of the handrails. Therefore, the objective identification module of the home environment does not output abstract scene labels, but rather project-level quantitative features that can be directly used to calculate the environment adaptation score. This allows for a more accurate prediction of the next stage of more suitable rehabilitation training content after the user completes functional assessment through wearable devices and machine vision, and combines this with the training performance in the previous stage of the intelligent rehabilitation training system.

[0030] In the specific implementation of step S1012, one embodiment is as follows: the multi-dimensional features include at least breathing features and movement features; The multi-dimensional features are obtained by performing the corresponding feature extraction method, including: A1. Calculate the dense optical flow displacement amplitude or inter-frame displacement sequence between adjacent frames of each region of interest in the video data of the multi-source heterogeneous data. A2. Process the displacement amplitude or inter-frame displacement sequence to obtain an estimated visible light respiration rate, which is then used as a respiration feature; A3. Combining the temporal sequence of key points on the human body, extract at least one preset multiple action sub-features as action features.

[0031] In steps A1-A3, this application first performs preprocessing on the visible light patient video data, including: brightness compensation, blur detection, overexposure detection, underexposure detection, and tracking stability assessment of the video frames; based on face detection bounding boxes... And / or human body key points, establish multiple regions of interest (ROIs) in the forehead, left cheek, right cheek, upper chest, left shoulder, and right shoulder. i ( t ), represented as: ROI i ( t )=( x f ( t )+ α i w f ( t ), y f ( t )+ β i h f ( t ), γi w f ( t ), δ i h f ( t )); Among them, Region of Interest (ROI) i ( t ) is the first i The position and size of a region of interest in frame t, i.e., the x and y coordinates of the top left corner of the rectangle, as well as its width and height, are used to distinguish multiple ROIs established on the forehead, left cheek, right cheek, upper chest, left shoulder, right shoulder, etc. t To indicate the first in the video t frame; α i , β i Indicates the first i The normalized offset of the region of interest relative to the top left corner of the face bounding box. γ i , δ i This indicates the aspect ratio of the area. This sensing area can reduce the risk of distortion when a single area is blocked, shadowed, moved by the head, or affected by localized reflections. x f ( t ) represents the x-coordinate (x-direction reference position) of the top-left corner of the face detection box in frame t. y f ( t ) represents the ordinate (y-axis reference position) of the top left corner of the face detection box in frame t. h f ( t ) represents the height of the face detection bounding box in frame t. ，w f ( t ) represents the width of the face detection bounding box in frame t, where the subscript... f The face frame is defined; based on the above formula, the effect of adaptively following the movement and scaling of the face frame is achieved even when the head moves or the frame is scaled.

[0032] Frames or regions of interest that do not meet the quality standards are downweighted or removed. After preprocessing, frames or regions of interest that meet the quality standards are processed as follows: For regions such as the forehead, left cheek, right cheek, and upper chest, a dense optical flow algorithm is used to calculate the inter-frame displacement amplitude or inter-frame displacement sequence. Specifically, for each ROI... i ( t ), calculate the dense optical flow between adjacent frames to obtain the horizontal displacement. With vertical displacement And construct the average motion amplitude of this region: ; in M i ( t ) represents the i-th region of interest in ... t Average motion amplitude at time; |ROI i ( t | indicates the number of pixels in the region. M i ( t This reflects the subtle displacement changes in the region caused by breathing and movement, thus obtaining the micro-motion signals of each region of interest over time. These micro-motion signals are generated by the chest rise and fall and subtle facial vibrations caused by breathing. To suppress interference, the displacement sequence is detrended and bandpass filtered to retain the typical breathing frequency band (approximately 0.1–0.5 Hz). Subsequently, spectral peak analysis is used to search for the frequency corresponding to the maximum peak value through power spectrum estimation. After multi-ROI signal-to-noise ratio weighted fusion, a robust visible light respiratory rate estimate is output and used as a respiratory feature. Simultaneously, a temporal convolutional model can be introduced, inputting the filtered multi-channel displacement sequence into the network to regress the respiratory waveform and instantaneous respiratory rate end-to-end. Specifically: , where RR cnn The respiratory rate estimate output by the temporal convolutional model, RR spec = 60 × f peak , f peak The frequency is the main peak frequency in the frequency domain. ω cnn The fusion weights are between 0 and 1. Kalman smoothing is used to further improve robustness to lighting and motion disturbances. This application also extracts motion features based on temporal data of human key points, specifically: using a posture estimation algorithm to extract the coordinates of key points such as shoulders, elbows, wrists, hips, knees, and ankles; analyzing kinematic parameters in continuous frames; for sit-stand training, detecting the turning point between sitting and standing by the change in vertical height of the hip and knee center points, calculating the time to complete a single sit-stand exercise, and accumulating the number of repetitions; the rhythm of stepping in place is obtained by zero-crossing detection of ankle joint relative displacement or autocorrelation analysis to obtain step frequency and its variability; gait stability is extracted by the lateral swing amplitude and stride width variation coefficient of the trunk center of mass (such as the midpoint of the two hips) to reflect dynamic balance ability; trunk swing amplitude is extracted by the tilt time sequence of the shoulder line midpoint relative to the vertical direction, calculating the maximum tilt angle and root mean square value; the range of motion of the upper limbs and chest is calculated by the shoulder-elbow-wrist three-point calculation of shoulder flexion and abduction angles, and by the relative displacement of the sternum and acromion to assess chest expansion, recording the maximum range of motion and symmetry during training. These motor characteristics collectively quantify a patient's motor ability, balance function, and training execution quality.

[0033] Brightness compensation is used to improve the adaptability of ordinary mobile phone videos under different lighting conditions by performing adaptive brightness compensation on each frame. Let the average brightness of the t-th frame be... Then the brightness compensation factor γ t Defined as: ;in, a Let be the brightness scaling constant, and ε be a small constant to prevent the denominator from being zero. The compensated image of frame t is denoted as... I t γ .

[0034] In a specific implementation of step S1012, one embodiment is as follows: the multi-dimensional features include at least thermal imaging respiratory features; The multi-dimensional features are obtained by performing the corresponding feature extraction method, including: B1. Target detection is performed on the nasal wing and / or nostril region of the thermal imaging data in the multi-source heterogeneous data, and region tracking is performed between adjacent detection frames; B2. Extract the temperature sequence of the coldest pixel in the tracked area, extract the thermal imaging respiration rate estimate from the coldest pixel temperature sequence, and use it as the thermal imaging respiration feature.

[0035] In steps B1-B2, this application first needs to perform small target detection on the nasal wing and / or nostril region in the thermal imaging video. To reduce the computational power consumption of the mobile phone, the detector is set to run once every N frames, and the ROI position is predicted by a Kalman tracker in the intermediate frames. This strategy can balance real-time performance and continuity. Because the airflow temperature in the nasal cavity is low during inhalation, this area appears as a local low-temperature patch in the thermal image. Temperature gradient segmentation or a lightweight detection network can be used for localization. To cope with the natural shaking of the user's head, Kalman tracking or optical flow tracking is introduced between adjacent frames to continuously update the position of the region of interest and ensure the stability of the sampling area. Let the nasal wing ROI in the thermal image of frame t be ROI. th ( t The temperature matrix is T t ( x , y Then, extract the temperature value of the coldest pixel in that region frame by frame: ; Here, α is the sensitivity coefficient, which is based on the temperature sequence of the coldest pixel, forming the temperature value of the coldest pixel. The coldest pixel directly reflects the cooling intensity of the nasal airflow on the mucosa; the temperature drops sharply during inhalation and rises during exhalation, making it more sensitive to changes in respiratory airflow than the regional average temperature. After obtaining the cold pixel temperature sequence, an adaptive threshold θ(t) = α×MAD(v(t)) +ε is set by the median absolute deviation of the temperature values ​​within a local window. A sliding window is applied to the cold pixel temperature sequence, and the median absolute deviation of the temperature values ​​within the window is calculated. This is then multiplied by an appropriate coefficient to obtain dynamic high and low thresholds, thereby overcoming the influence of individual baseline differences and ambient temperature drift. The dynamic high and low thresholds are input into a hysteresis state machine for inhalation, exhalation, and hold state recognition. The hysteresis mechanism can effectively filter out small noise fluctuations and prevent frequent phase jumps. ; in φ th ( t ) represents the respiratory phase marker at time t; φ th ( t ) = +1 indicates the expiratory phase; φ th ( t ) = 1 indicates the inspiratory phase; φ th ( t A value of 0 indicates a transition phase or hold phase. This means: when the temperature drop exceeds a high threshold, the breath enters the inspiratory phase; when the temperature rises and exceeds a low threshold, the breath switches to the expiratory phase; otherwise, it is a transition phase. The time interval between the starting points of adjacent phases of the same type is used to determine the transition phase. This allows us to obtain an estimate of the thermal imaging respiration rate, which can then be used as a thermal imaging respiration feature. ; Based on the identified continuous respiratory cycles, the coefficients of variation of thermal imaging respiratory rate and inspiratory and expiratory durations are statistically obtained, thereby quantifying the stability of respiratory rhythm.

[0036] In step S102, after obtaining multi-dimensional features including physiological state characteristics, movement characteristics, respiratory state characteristics, thermal imaging respiratory characteristics, respiratory flow characteristics, leakage state characteristics, environmental resource characteristics, and environmental risk characteristics, this application calculates multiple functional domain scores for pre-divided functional domains based on the multi-dimensional features. The functional domains include at least the respiratory function domain, activity tolerance domain, mobility domain, upper limb shoulder and chest movement domain, and balance and safety domain. Based on the multi-dimensional features and the multiple functional domain scores, multi-dimensional parameter values ​​for each candidate rehabilitation training item in the user's next stage are collaboratively calculated. This includes training needs score, environment adaptation score, training performance score, safety gating factor, and postoperative stage coefficient. The postoperative stage coefficient is determined based on whether the user is in the early postoperative period, transition period, or stable improvement period after lung cancer surgery. It calculates the user's performance in each candidate rehabilitation training program in terms of training needs score, environment adaptation score, training performance score, and safety gating factor to help select the user's rehabilitation training program for the next stage. When any parameter value does not meet the threshold condition of the corresponding candidate rehabilitation training program, the corresponding candidate rehabilitation training program is prohibited from being recommended, downgraded, or replaced with a low-intensity candidate rehabilitation training program.

[0037] In a specific implementation of step S102, one embodiment involves calculating multiple functional domain scores based on the multi-dimensional features, including: S10211. For multiple functional domains, extract corresponding features from the multi-dimensional features and normalize the features; S10212. The normalized features are processed by a pre-set functional domain calculation model to obtain the corresponding functional domain score.

[0038] In steps S10211-S10212, this application extracts corresponding features from the multi-dimensional features for the divided respiratory function domain, activity tolerance domain, mobility domain, upper limb shoulder and chest movement domain, and balance and safety domain. For example, the features corresponding to the respiratory function domain are respiratory state features, thermal imaging respiratory features, respiratory flow features, etc., and normalizes the features. After normalization, the normalized features are processed by a pre-set functional domain calculation model to obtain the corresponding functional domain score. Then, for the d-th functional domain, its score is defined. F d The result is calculated by the functional domain computation model, which includes the normalized feature input mapping model or the rule model, as shown below: Fd = σ (∑ kμ dk z dk+ bd ); in, For Sigmoid mapping, z dk This is the k-th normalized feature of the functional domain. μ dk For the corresponding weights, b d For bias terms, F d The value ranges from 0 to 1, and the larger the value, the better the state of the functional domain.

[0039] In the specific implementation of step S102, another embodiment is as follows: based on the multi-dimensional features and multiple functional domain scores, the multi-dimensional parameter values ​​of each candidate rehabilitation training item for the user's next stage are calculated collaboratively, including: S10221. Set multi-dimensional standard parameters based on candidate rehabilitation training programs, user's physical condition, current family environment, actual skill mastery, and adaptation status. S10222. Perform calculations on the multi-dimensional features and the scores of multiple functional domains according to the multi-dimensional standard parameters to obtain the multi-dimensional parameter values.

[0040] In steps S10221-S10222, this application sets multi-dimensional standard parameters based on candidate rehabilitation training projects, user's physical condition, current family environment, actual skill mastery, and adaptation status. These parameters include training demand score, environmental adaptation score, training performance score, safety gating factor, and post-operative stage coefficient. The multi-dimensional features and multiple functional domain scores are then calculated using the corresponding multi-dimensional standard parameters to obtain the multi-dimensional parameter values. For example, the training demand score is calculated based on the preset correspondence between each candidate rehabilitation training project and the multiple functional domains, according to the degree of functional domain deficit. The higher the degree of correspondence between the candidate rehabilitation training project and the deficit functional domain, the higher its training demand score. The higher the training demand score, the better. The training performance score is calculated based on at least three of the following within a preset statistical window: training completion rate, movement quality attainment rate, improvement trend, symptom burden, training interruption rate, and recovery time. The improvement trend is determined by the change in target indicators between the current statistical window and at least one previous statistical window. The environmental fit score is calculated based on the positive contribution of the environmental resource characteristics and the negative contribution of the environmental risk characteristics. The safety gating factor is determined based on at least one of the following during training: minimum blood oxygen saturation, heart rate change or peak heart rate, abnormal respiratory events, balance risk indicators, leakage state characteristics, dyspnea score, fatigue score, and pain score. (For candidate training items...) T j Define its training requirement score N j : ; in, N j Indicates training project T j The response weight for the defect in the d-th functional domain. η jd Indicates training project T j Response weights for functional domain d deficiencies. The more pronounced the functional domain deficiency, the higher the required score for the corresponding training item. For example, if the scores for the activity endurance domain and respiratory function domain are low, the required scores for walking endurance training, breathing pattern training, and respiratory muscle training will increase; if the score for the upper limb shoulder and chest range of motion domain is low, the required scores for affected side shoulder and chest range of motion training and thoracic movement training will increase.

[0041] In order to incorporate "how well you've been training lately" into the recommendation logic, this application defines training projects. T j Training performance score within the most recent statistical window P j : ; in: P j For training projects T j Training performance score; A j For training completion rate; Q j To achieve the required rate of action quality; U j This reflects a recent trend of improvement. B j As an indicator of symptom burden; I j β1 represents the training interruption rate; β1-β5 represent the weight coefficients. B j It consists of indicators such as difficulty breathing, fatigue, pain, and abnormal cough. P j The larger the value, the better the training program performed in the previous stage and the more solid the foundation for advanced learning. P j The smaller the value, the greater the current execution resistance or the heavier the symptom burden. For each candidate training item... T j The system sets security gating factors. G j : G j ={1, When minimum blood oxygen , Maximum heart rate increment , Abnormal respiratory indicators , Balancing risks , Attachment leakage status and When the subjective symptom burden is within the allowable range ;0, Other situations}; In the formula, G j For training projects T j The security gating factor. When any of the following conditions occur... G j = 0: Blood oxygen levels are below the corresponding threshold, heart rate recovery is abnormal, persistent abnormal breathing patterns occur during training, there is significant leakage in differential pressure accessories, gait or balance risks exceed limits, and subjective symptom scores exceed the allowable upper limit. The safety gating factor is used to avoid directly recommending corresponding training programs when there is only "training need" but no "safety feasibility".

[0042] In step S103, based on the training requirement score, environment adaptation score, training performance score, postoperative stage coefficient, and safety gating factor included in the multi-dimensional parameter values, this application calculates the predicted adaptation value of each candidate rehabilitation training item in the next stage, and performs this calculation for each candidate training item. T j Define the next stage fitness prediction value S j : ; in: S j For candidate training projects T j The next stage of adaptation prediction; G j For security gating factors; N j The score is for training requirements; E j Environmental adaptation score; P j This is a score for training performance; C j ph This refers to the postoperative stage coefficient; λ 1 to λ 4 represents the weighting coefficient. When At that time, training program T j To be included in the next stage of the recommended training set; when multiple items meet the conditions, this application will be processed according to... S j The output is sorted from highest to lowest. It also outputs the next stage of training content where the fitness prediction value meets preset conditions, and generates training dose parameters corresponding to the next stage of training content. Let the current dose be... Dose j curThen the next stage dose Dose j next for: ; In the formula, Dose j next For training projects T j The next phase of training dosage; Dose j cur This is the current training dose; P j This is a score for training performance; B j As an indicator of symptom burden; E j Environmental adaptation score; κ 1. κ 2 and κ 3 represents the dose adjustment weight; D j min and D j max These are the minimum and maximum doses, respectively; clip( x , l , u The expression indicates that x is restricted to the interval [l, u]. This formula means that the better the performance and the more suitable the environment in the previous stage, the dosage can be appropriately increased; the higher the symptom burden, the lower the dosage should be accordingly. The training dosage parameters include at least one or more of the following: training duration, number of repetitions, number of sets, target distance, target speed, and target number of steps. These training dosage parameters are updated within preset upper and lower limits based on the current training dosage, training performance score, symptom burden, and environment suitability score.

[0043] In the specific implementation of step S103, one embodiment is as follows: Figure 3 As shown, the user's next stage of training projects and corresponding training dose parameters are obtained, including: S1031. Sort each candidate rehabilitation training item based on the fit prediction value to select the user's next stage training item. S1032. Based on the user's current training performance, current symptoms, and current home environment, dynamically determine the training dosage parameters corresponding to the next stage of training.

[0044] In steps S1031-S1032, this application first constructs a candidate rehabilitation training item library, wherein the candidate rehabilitation training item library The training program includes at least one or more of the following: breathing pattern training, pursed-lip breathing training, respiratory muscle training, effective coughing or expectoration training, chest wall movement training, affected side shoulder and chest range of motion training, sit-stand training, indoor walking training, balance training, step training, and low-intensity endurance training. For each candidate training item, a fitness prediction model is used for quantitative scoring. This model comprehensively considers the patient's current multidimensional state: physiological capabilities are assessed by integrating indicators such as visible light and thermal imaging respiratory rate, respiratory rhythm stability, and heart rate recovery rate; motor function is assessed by incorporating movement characteristics such as sit-stand completion time, gait stability, trunk swing amplitude, and shoulder joint range of motion; training history is assessed by statistically analyzing the completion rate, movement quality scores, and fatigue recovery curves of similar previous training items; symptom dimensions are assessed by collecting patient-reported scores for current symptoms such as pain, shortness of breath, and fatigue; and environmental fitness is assessed by evaluating whether training safety conditions are met based on environmental identification results such as home space dimensions, floor slip resistance, and the presence or absence of support structures (e.g., armrests, chair backs). After weighted fusion or lightweight classifier processing, each input item is output as a suitability score for the current context. A higher score indicates greater training benefit and manageable risk. All candidate items are sorted in descending order of score, and the top-ranked item is selected as the recommended training item for the next stage. Based on the user's current training performance, if the movement is performed correctly, heart rate variability recovers well, and there is no significant compensation, the number of repetitions or duration is appropriately increased; if movement deformities, excessive fatigue, or decreased blood oxygen levels occur, the intensity is reduced. Simultaneously, current symptoms are considered; for example, mild pain is avoided, and the frequency of rest intervals is increased for shortness of breath. Furthermore, the home environment is taken into account; narrow spaces limit stride length and walking distance, and unsupported conditions reduce the difficulty of balance training. Finally, individualized dosage parameters are generated, including the number of sets, repetitions, duration, target heart rate zone, and safe termination threshold, achieving accurate and safe dynamic recommendations for training prescriptions.

[0045] This application can simultaneously utilize both contact and non-contact data, avoiding instability in evaluation caused by single sensor failure, occlusion, or changes in lighting. Multi-ROI optical flow, thermal imaging nose wing tracking, and differential pressure flow estimation provide compensation information under different conditions. It can objectively identify home training environments, no longer treating the home scene as a black box, thus unifying "what training is suitable" with "what training can be safely done at home," improving the executability of the recommendations. It can convert the training performance of the previous stage into a calculable training performance score. P j This enables the method provided in this application to have individualized advancement capabilities, rather than mechanically advancing according to the number of days after surgery; it can be controlled by safety gating factors. G jIt can constrain hypoxia, abnormal heart rate, balance risk, and leakage status, thereby reducing the risks caused by inappropriate training recommendations; it can output the training content, priority and training dosage for the next stage, and is suitable for seamless integration with existing intelligent rehabilitation training systems to form a complete home rehabilitation closed loop.

[0046] The following is an example of this application: The user is a lung cancer surgery patient undergoing home rehabilitation. This application includes a smartwatch, a clip-on or wrist-mounted pulse oximeter, a mobile phone camera, optional thermal imaging accessories, optional differential pressure respiratory training accessories, and a smart rehabilitation training app. The user completes a home scene scan and a standardized functional assessment action capture once a day before training. First, the mobile phone camera acquires videos of the patient in a frontal sitting position, standing position, sitting-standing position, stepping in place, and the training area. The visible light module establishes ROIs on the forehead, cheeks, upper chest, and shoulders, extracts optical flow time series, and calculates RR. vis The system detects chest rise and fall amplitude surrogate values, sit-stand rhythm, shoulder and chest range of motion, and gait stability features. If insufficient lighting or low image quality is detected, a thermal imaging accessory is activated to detect the nasal ROI, extract the temperature curve of the coldest pixel, and obtain the RR based on the MAD adaptive threshold and hysteresis state machine. th And the stability of the respiratory rhythm. If the user performs respiratory muscle training on the same day, the differential pressure breathing training accessory will simultaneously record ΔP and V. b ,estimate Q resp,e and according to I leak The system determines whether the training accessories are leaking air or improperly worn. Simultaneously, it performs object detection and segmentation in a home setting, identifying walkable areas, obstacles, chairs, backrests, handrails, carpets / puddles, and other floor hazards, as well as stairs / steps. This yields... e space , e chair , h obs Environmental metrics were evaluated, and an environment fit score was calculated for each candidate training item. E j Next, based on the training logs from the past 7 days, the completion rate of each training item was calculated. A j quality of movement Q j Recent improvement trend U j Symptom burden B j and interruption rate I j And thus obtain P j .

[0047] Assuming a user has a low activity tolerance score, a moderately low respiratory function score, a fair balance safety score, and a home environment with usable walking space and suitable seating, but no reliable handrails on the stairs, the system might obtain the following results: Training demand score for indoor walking training. N j High environmental adaptability score E j High, training performance score P j Medium and security gating passed, therefore the next stage of fitness prediction value S j Higher; sitting and standing training S j The score is also relatively high; while step training, although in demand, has a lower score due to environmental adaptation. E j Low or safety gate factor G j It fails and is therefore not included in the next stage of the recommendation set. The final output is a combination of "breathing pattern training, chest wall movement training, affected side shoulder and chest range of motion training, indoor walking training, and sitting-standing training" as the recommended content for the next stage, and automatically provides the duration, number of repetitions or distance increment for each training exercise.

[0048] Example 2 This application also provides an assessment device for home rehabilitation training after lung cancer surgery, such as... Figure 4 The diagram shows a block diagram of an assessment device for home rehabilitation training after lung cancer surgery. This device performs functions corresponding to the steps of the aforementioned assessment method for home rehabilitation training after lung cancer surgery executed on a terminal device. The device can be understood as a server component including a processor. The assessment device for home rehabilitation training after lung cancer surgery described in this application includes: The acquisition module 401 is used to acquire multi-source heterogeneous data of lung cancer postoperative users in the home rehabilitation training stage within a preset assessment window based on wearable devices and mobile phone cameras, and extract multi-dimensional features from the multi-source heterogeneous data. The scoring module 402 is used to calculate multiple functional domain scores of multiple pre-divided functional domains based on the multi-dimensional features, and to collaboratively calculate the multi-dimensional parameter values ​​of each candidate rehabilitation training item for the user's next stage based on the multi-dimensional features and the multiple functional domain scores. The calculation module 403 is used to calculate the predicted fit value of each candidate rehabilitation training item in the next stage based on the multi-dimensional parameter values, so as to obtain the next stage training items and corresponding training dosage parameters for the user in the home rehabilitation training stage.

[0049] In one feasible implementation, the acquisition module includes: Pre-divide standard multi-source heterogeneous data into multiple standard data types; different standard data types correspond to different feature extraction methods; Identify the standard data type corresponding to the user's multi-source heterogeneous data, and perform the corresponding feature extraction method to obtain the multi-dimensional features.

[0050] In one feasible implementation, the acquisition module further includes: Calculate the dense optical flow displacement amplitude or inter-frame displacement sequence between adjacent frames of each region of interest in the video data of the multi-source heterogeneous data. The displacement amplitude or inter-frame displacement sequence is processed to obtain an estimated visible light respiration rate, which is then used as a respiration feature. By combining the temporal sequence of key points on the human body, at least one preset multiple action sub-features are extracted as action features.

[0051] In one feasible implementation, the acquisition module also includes: Target detection is performed on the nasal wing and / or nostril region of the thermal imaging data in the multi-source heterogeneous data, and region tracking is performed between adjacent detection frames; The temperature sequence of the coldest pixel in the tracked area is extracted, and the thermal imaging respiration rate estimate is extracted from the temperature sequence of the coldest pixel and used as the thermal imaging respiration feature.

[0052] In one feasible implementation, the scoring module includes: Multi-dimensional standard parameters are set based on candidate rehabilitation training programs, user's physical condition, current family environment, actual skill mastery, and adaptation status. The multi-dimensional parameter values ​​are obtained by performing calculations on the multi-dimensional features and the scores of multiple functional domains according to the corresponding multi-dimensional standard parameters.

[0053] In one feasible implementation, the computing module includes: Candidate rehabilitation training programs are ranked based on their fit prediction values ​​to select the user's next stage of training. Based on the user's current training performance, current symptoms, and current home environment, the training dosage parameters corresponding to the next stage of training are dynamically determined.

[0054] In one feasible implementation, the acquisition module further includes: For multiple functional domains, corresponding features are extracted from the multi-dimensional features, and the features are normalized. The normalized features are processed by a pre-set functional domain calculation model to obtain the corresponding functional domain score.

[0055] Example 3 This application also provides an electronic device, such as Figure 5 As shown, it includes: a processor 501, a memory 502, and a bus 503. The memory 502 stores machine-readable instructions that can be executed by the processor 501. When the electronic device is running, the processor 501 and the memory 502 communicate through the bus 503. When the machine-readable instructions are executed by the processor 501, they perform the steps of any one of the assessment methods for home rehabilitation training after lung cancer surgery.

[0056] Example 4 This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of any one of the assessment methods for postoperative home rehabilitation training for lung cancer.

[0057] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0058] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0059] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0060] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a platform server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0061] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An assessment method for home rehabilitation training after lung cancer surgery, characterized in that, The method includes: Based on wearable devices and mobile phone cameras, multi-source heterogeneous data of lung cancer postoperative users in the home rehabilitation training stage are acquired within a preset assessment window, and multi-dimensional features are extracted from the multi-source heterogeneous data. Based on the multi-dimensional features, multiple functional domain scores of the pre-divided functional domains are calculated, and based on the multi-dimensional features and multiple functional domain scores, multi-dimensional parameter values ​​of each candidate rehabilitation training item for the user's next stage are calculated collaboratively. Based on the multi-dimensional parameter values, the predicted fit of each candidate rehabilitation training item in the next stage is calculated to obtain the next stage training items and corresponding training dosage parameters for the user in the home rehabilitation training stage.

2. The method according to claim 1, characterized in that, Extracting multi-dimensional features from the multi-source heterogeneous data, including: Pre-divide standard multi-source heterogeneous data into multiple standard data types; different standard data types correspond to different feature extraction methods; Identify the standard data type corresponding to the user's multi-source heterogeneous data, and perform the corresponding feature extraction method to obtain the multi-dimensional features.

3. The method according to claim 2, characterized in that, The multidimensional features include at least respiratory features and movement features. The multi-dimensional features are obtained by performing the corresponding feature extraction method, including: Calculate the dense optical flow displacement amplitude or inter-frame displacement sequence between adjacent frames of each region of interest in the video data of the multi-source heterogeneous data. The displacement amplitude or inter-frame displacement sequence is processed to obtain an estimated visible light respiration rate, which is then used as a respiration feature. By combining the temporal sequence of key points on the human body, at least one preset multiple action sub-features are extracted as action features.

4. The method according to claim 2, characterized in that, The multidimensional features include at least thermal imaging respiratory features; The multi-dimensional features are obtained by performing the corresponding feature extraction method, including: Target detection is performed on the nasal wing and / or nostril region of the thermal imaging data in the multi-source heterogeneous data, and region tracking is performed between adjacent detection frames; The temperature sequence of the coldest pixel in the tracked area is extracted, and the thermal imaging respiration rate estimate is extracted from the temperature sequence of the coldest pixel and used as the thermal imaging respiration feature.

5. The method according to claim 1, characterized in that, Based on the aforementioned multi-dimensional features and multiple functional domain scores, the multi-dimensional parameter values ​​of each candidate rehabilitation training item for the user's next stage are calculated collaboratively, including: Multi-dimensional standard parameters are set based on candidate rehabilitation training programs, user's physical condition, current family environment, actual skill mastery, and adaptation status. The multi-dimensional parameter values ​​are obtained by performing calculations on the multi-dimensional features and the scores of multiple functional domains according to the corresponding multi-dimensional standard parameters.

6. The method according to claim 1, characterized in that, Obtain the user's next stage of training projects and corresponding training dose parameters, including: Candidate rehabilitation training programs are ranked based on their fit prediction values ​​to select the user's next stage of training. Based on the user's current training performance, current symptoms, and current home environment, the training dosage parameters corresponding to the next stage of training are dynamically determined.

7. The method according to claim 1, characterized in that, Based on the multi-dimensional features, multiple functional domain scores are calculated for pre-divided functional domains, including: For multiple functional domains, corresponding features are extracted from the multi-dimensional features, and the features are normalized. The normalized features are processed by a pre-set functional domain calculation model to obtain the corresponding functional domain score.

8. An assessment device for home rehabilitation training after lung cancer surgery, characterized in that, The device includes: The acquisition module is used to acquire multi-source heterogeneous data of lung cancer postoperative users in the home rehabilitation training stage within a preset assessment window based on wearable devices and mobile phone cameras, and extract multi-dimensional features from the multi-source heterogeneous data. The scoring module is used to calculate multiple functional domain scores for multiple pre-divided functional domains based on the multi-dimensional features, and to collaboratively calculate the multi-dimensional parameter values ​​of each candidate rehabilitation training item for the user's next stage based on the multi-dimensional features and the multiple functional domain scores. The calculation module is used to calculate the predicted fit of each candidate rehabilitation training item in the next stage based on the multi-dimensional parameter values, so as to obtain the next stage training items and corresponding training dosage parameters for the user in the home rehabilitation training stage.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of an assessment method for home rehabilitation training after lung cancer surgery as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of an assessment method for postoperative home rehabilitation training for lung cancer as described in any one of claims 1 to 7.

Images