Non-contact sarcopenia risk assessment system and method based on multi-modal thermal imaging

Abstract

The invention discloses a non-contact sarcopenia risk assessment system and method based on multi-mode thermal imaging, and the system comprises a data collection module which is used for collecting a thermal imaging video and an RGB video; the region tracking module is used for carrying out key point detection and region-of-interest positioning on the thermal imaging video and the RGB video to obtain first quadriceps femoris region-of-interest data, and carrying out quality inspection and motion compensation on the first quadriceps femoris region-of-interest data to obtain second quadriceps femoris region-of-interest data; the feature extraction module is used for performing time sequence feature extraction and reliability calculation on the second quadriceps femoris region-of-interest data to obtain a time sequence feature vector and a reliability index; and the risk assessment module is used for inputting the time sequence feature vector into a plurality of sarcopenia risk prediction models to obtain a sarcopenia risk assessment result. The sarcopenia risk assessment accuracy is improved, the risks of radiation and contact are avoided, and the sarcopenia risk assessment method can be widely applied to the technical field of data processing.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a non-contact sarcopenia risk assessment system and method based on multimodal thermal imaging. Background Technology

[0002] Existing methods for identifying sarcopenia include dual-energy X-ray absorptiometry (DEXA), bioelectrical impedance analysis (BIA), ultrasound, and infrared thermography. DEXA measures muscle mass (ASMI index) using X-rays; however, its high equipment cost and radiation risks limit its application in high-frequency screening. BIA estimates muscle mass based on impedance measurements using surface electrodes, but this method has significant errors in obese individuals and requires direct skin contact, making it less acceptable to some elderly individuals. Ultrasound is primarily used to measure muscle thickness, but this method is highly dependent on operator experience and cannot dynamically assess metabolic activity, potentially leading to significant errors. Traditional infrared thermography relies only on single-frame temperature thresholds and therefore cannot capture the dynamic thermodynamic characteristics after exercise. Summary of the Invention

[0003] To address the aforementioned technical problems, the purpose of this application is to provide a low-cost, efficient, and highly accurate non-contact sarcopenia risk assessment system and method based on multimodal thermal imaging.

[0004] To achieve the above objectives, one aspect of this application proposes a non-contact sarcopenia risk assessment system based on multimodal thermal imaging, comprising:

[0005] The data acquisition module is used to acquire thermal imaging videos and RGB videos of the subject at multiple motion phases;

[0006] The region tracking module is used to perform key point detection and region of interest localization on the thermal imaging video and the RGB video to obtain the first quadriceps femoris region of interest data, and then to perform quality checks and motion compensation on the first quadriceps femoris region of interest data to obtain the second quadriceps femoris region of interest data.

[0007] The feature extraction module is used to extract temporal features from the second quadriceps muscle region of interest data to obtain a temporal feature vector, and to calculate the reliability of the second quadriceps muscle region of interest data to obtain a reliability index.

[0008] The risk assessment module is used to input the time-series feature vector into multiple pre-trained sarcopenia risk prediction models to obtain sarcopenia risk prediction values, and to obtain sarcopenia risk assessment results based on the sarcopenia risk prediction values ​​and the reliability index.

[0009] The time-series feature vector includes the baseline temperature mean, maximum temperature rise, recovery half-life, metabolic recovery index, and thermo-metabolic efficiency.

[0010] In some embodiments, the motion phase includes a baseline phase, a motion phase, and a recovery phase, and the region tracking module includes:

[0011] The key point detection module is used to calibrate the RGB video and the thermal imaging video, and then perform key point detection on the thermal imaging video corresponding to the movement period based on the calibrated RGB video to obtain the key points of the lower limbs.

[0012] The boundary segmentation module is used to segment the quadriceps femoris boundary in the thermal imaging video based on the key points of the lower limbs to obtain the optimal region of interest.

[0013] The boundary alignment module is used to align the optimal region of interest with the thermal imaging videos corresponding to the baseline period and the recovery period to obtain the first quadriceps region of interest data corresponding to each of the motion stages.

[0014] In some embodiments, the motion phase includes a baseline phase, a motion phase, and a recovery phase, and the region tracking module includes:

[0015] The quality inspection module is used to detect abnormal frames in the first quadriceps muscle region of interest data corresponding to each of the movement stages, and to perform spatial alignment verification on the first quadriceps muscle region of interest data corresponding to each of the movement stages to obtain abnormal situations, and then repair the abnormal situations.

[0016] The motion compensation module is used to perform thermal stability correction on the repaired region of interest data of the first quadriceps femoris corresponding to the baseline period, perform global displacement compensation on the repaired region of interest data of the first quadriceps femoris corresponding to the exercise period, and perform static compensation on the repaired region of interest data of the first quadriceps femoris corresponding to the recovery period, so as to obtain the region of interest data of the second quadriceps femoris corresponding to each exercise stage.

[0017] In some embodiments, the feature extraction module includes:

[0018] The statistical temperature calculation module is used to extract all pixel temperature values ​​in the second quadriceps muscle region of interest data, construct a temperature matrix, and then calculate the statistical temperature of the region of interest based on the temperature matrix.

[0019] The baseline feature extraction module is used to determine a first sampling window, and within the first sampling window, calculate the mean baseline temperature based on the statistical temperature of the region of interest.

[0020] The motion period feature extraction module is used to determine the second sampling window. Within the second sampling window, the peak search is performed on the statistical temperature of the region of interest, and then the maximum temperature rise is calculated based on the peak search result and the baseline temperature mean.

[0021] The recovery period feature extraction module is used to determine the third sampling window. Within the third sampling window, the recovery half-life, recovery integrity, and recovery slope are calculated based on the statistical temperature of the region of interest, the mean baseline temperature, the peak search result, and the maximum temperature rise.

[0022] The feature index calculation module is used to calculate the metabolic recovery index based on the recovery integrity, and to calculate the thermal metabolic efficiency based on the maximum temperature rise and the recovery slope.

[0023] In some embodiments, the feature extraction module includes:

[0024] The data integrity assessment module is used to determine the missing proportion and the stability of the region of interest, and then calculate the data integrity score based on the missing proportion and the stability of the region of interest.

[0025] The signal-to-noise ratio (SNR) evaluation module is used to determine the baseline temperature stability and then calculate the SNR score based on the maximum temperature rise and the baseline temperature stability.

[0026] The fitting quality assessment module is used to determine the goodness of fit of the recovery curve, and then calculate the fitting quality score based on the goodness of fit of the recovery curve and the baseline temperature stability.

[0027] The physiological rationality assessment module is used to assess the range of the baseline temperature mean, the maximum temperature rise, and the recovery half-life, and then calculate the physiological rationality score based on the range assessment results.

[0028] The reliability index calculation module is used to determine the weight values, and based on the weight values, to perform a weighted summation of the data integrity score, the signal-to-noise ratio score, the fitting quality score, and the physiological rationality score to obtain the reliability index.

[0029] In some embodiments, the non-contact sarcopenia risk assessment system based on multimodal thermal imaging further includes a parameter correction module, the parameter correction module comprising:

[0030] The parameter acquisition module is used to acquire the subject's age, gender, body mass index, and medical history, and to collect environmental temperature and humidity data.

[0031] The correction coefficient determination module is used to determine a first correction coefficient based on the age and the gender, a second correction coefficient based on the body mass index, a third correction coefficient based on the medical history, and a fourth correction coefficient based on the environmental temperature and humidity data.

[0032] The feature vector correction module is used to correct the time series feature vector according to the first correction coefficient, the second correction coefficient, the third correction coefficient and the fourth correction coefficient to obtain the corrected time series feature vector.

[0033] In some embodiments, the sarcopenia risk prediction model includes a first risk prediction model, a second risk prediction model, and a third risk prediction model. The non-contact sarcopenia risk assessment system based on multimodal thermal imaging further includes a model training module, which includes:

[0034] The training data acquisition module is used to acquire a training dataset, which includes a number of training samples and labeled data corresponding to each training sample.

[0035] The training data processing module is used to extract features from the training samples to obtain sample feature vectors, and to collect the age, gender, body mass index, and medical history of the training samples to obtain extended features of the samples.

[0036] The first model training module is used to input the sample feature vector and the sample extended features into a preset gradient boosting decision tree model for training, and to optimize the parameters of the trained gradient boosting decision tree model according to the labeled data to obtain the first risk prediction model.

[0037] The second model training module is used to input the sample feature vector and the sample extended features into a preset random forest model for training, and to optimize the parameters of the trained random forest model according to the labeled data to obtain the second risk prediction model.

[0038] The third model training module is used to input the sample feature vector and the sample extended features into a preset support vector machine model for training, and to optimize the parameters of the trained support vector machine model based on the labeled data to obtain the third risk prediction model.

[0039] In some embodiments, the sarcopenia risk prediction model includes a first risk prediction model, a second risk prediction model, and a third risk prediction model, and the risk assessment module includes:

[0040] The first model prediction module is used to input the corrected time-series feature vector into the first risk prediction model and output the first sarcopenia probability value.

[0041] The second model prediction module is used to input the corrected time-series feature vector into the second risk prediction model and output the second sarcopenia probability value.

[0042] The third model prediction module is used to input the corrected time-series feature vector into the third risk prediction model and output the third sarcopenia probability value.

[0043] The risk prediction value determination module is used to perform a weighted summation of the first sarcopenia probability value, the second sarcopenia probability value, and the third sarcopenia probability value, and then determine the maximum value in the weighted summation result as the sarcopenia risk prediction value.

[0044] In some embodiments, the risk assessment module includes:

[0045] The confidence calculation module is used to calculate the confidence level based on the sarcopenia risk prediction value and the reliability index.

[0046] The risk assessment output module is used to output the sarcopenia risk assessment result based on the confidence level and the sarcopenia risk prediction value.

[0047] To achieve the above objectives, another aspect of this application proposes a non-contact sarcopenia risk assessment method based on multimodal thermal imaging, comprising the following steps:

[0048] Thermal and RGB videos of the subjects were acquired during multiple motion phases.

[0049] Key point detection and region of interest localization are performed on the thermal imaging video and the RGB video to obtain the first quadriceps femoris region of interest data. Then, the first quadriceps femoris region of interest data is subjected to quality inspection and motion compensation to obtain the second quadriceps femoris region of interest data.

[0050] Temporal features are extracted from the region of interest data of the second quadriceps femoris muscle to obtain a temporal feature vector, and the reliability of the region of interest data of the second quadriceps femoris muscle is calculated to obtain a reliability index.

[0051] The time-series feature vector is input into multiple pre-trained sarcopenia risk prediction models to obtain sarcopenia risk prediction values. Based on the sarcopenia risk prediction values ​​and the reliability index, the sarcopenia risk assessment result is obtained.

[0052] The time-series feature vector includes the baseline temperature mean, maximum temperature rise, recovery half-life, metabolic recovery index, and thermo-metabolic efficiency.

[0053] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0054] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method.

[0055] The beneficial effects of this application are as follows: The non-contact sarcopenia risk assessment system and method based on multimodal thermal imaging includes a data acquisition module, a region tracking module, a feature extraction module, and a risk assessment module. This application utilizes infrared thermal imaging technology to capture thermal imaging videos of subjects at multiple stages of movement. Based on these thermal imaging videos, baseline mean temperature, maximum temperature rise, recovery half-life, metabolic recovery index, and thermometabolic efficiency are extracted. By comprehensively assessing the neuromuscular synergistic efficiency through these extracted dynamic indicators, the accuracy of risk assessment is further improved. Then, the extracted baseline mean temperature, maximum temperature rise, recovery half-life, metabolic recovery index, and thermometabolic efficiency are input into multiple trained sarcopenia risk prediction models to automatically obtain sarcopenia risk assessment results. This avoids the risks of radiation and contact, greatly improves the processing efficiency of thermal imaging videos, and reduces the operational threshold and equipment usage costs. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments of this application are described below. It should be understood that the drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions in this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0057] Figure 1 This is a schematic diagram of the structure of a non-contact sarcopenia risk assessment system based on multimodal thermal imaging, provided in one embodiment of this application.

[0058] Figure 2 A schematic diagram of a three-stage standardized data acquisition process provided in one embodiment of this application;

[0059] Figure 3 This is a schematic diagram of a multi-level correction process provided in one embodiment of this application;

[0060] Figure 4 A flowchart illustrating the steps of a non-contact sarcopenia risk assessment method based on multimodal thermal imaging provided in an embodiment of this application;

[0061] Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0063] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0064] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0065] Existing methods for identifying sarcopenia include dual-energy X-ray absorptiometry (DEXA), bioelectrical impedance analysis (BIA), ultrasound, and infrared thermography. DEXA measures muscle mass (ASMI index) using X-rays; however, its high equipment cost and radiation risks limit its application in high-frequency screening. BIA estimates muscle mass based on impedance measurements using surface electrodes, but this method has significant errors in obese individuals and requires direct skin contact, making it less acceptable to some elderly individuals. Ultrasound is primarily used to measure muscle thickness, but this method is highly dependent on operator experience and cannot dynamically assess metabolic activity, potentially leading to significant errors. Traditional infrared thermography relies only on single-frame temperature thresholds and therefore cannot capture the dynamic thermodynamic characteristics after exercise.

[0066] In summary, existing methods for identifying sarcopenia have several shortcomings. First, regarding invasiveness and safety, dual-energy X-ray absorptiometry (DEXA) is unsuitable for frequent use due to radiation exposure issues, while bioelectrical impedance analysis (BIA) requires electrode placement, making it less acceptable to the elderly. Second, static assessments are significantly limited; current technologies can only provide muscle mass data and cannot reflect dynamic functional indicators such as neural drive efficiency and metabolic recovery rate. Furthermore, ultrasound testing is highly complex, requiring professionals to manually locate regions of interest (ROIs), while traditional thermal imaging lacks standardized action protocols, further affecting data accuracy and reliability. From a cost and accessibility perspective, DEXA equipment is expensive, making widespread adoption in communities difficult. As for the application of traditional dynamic infrared thermal imaging in muscle monitoring, it is limited to capturing only overall temperature changes after exercise, failing to quantify the temporal thermodynamic characteristics of specific muscle groups, and lacking standardized action protocols and correlation models related to sarcopenia.

[0067] In view of this, this application proposes a non-contact sarcopenia risk assessment system based on multimodal thermal imaging, including a data acquisition module, a region tracking module, a feature extraction module, and a risk assessment module. This application utilizes infrared thermal imaging technology to capture thermal imaging videos of subjects at multiple stages of movement. Based on these thermal imaging videos, baseline mean temperature, maximum temperature rise, recovery half-life, metabolic recovery index, and thermometabolic efficiency are extracted. By comprehensively assessing the neuromuscular synergistic efficiency through these extracted dynamic indicators, the accuracy of risk assessment is further improved. Then, the extracted baseline mean temperature, maximum temperature rise, recovery half-life, metabolic recovery index, and thermometabolic efficiency are input into multiple trained sarcopenia risk prediction models to automatically obtain sarcopenia risk assessment results. This avoids the risks of radiation and contact, greatly improves the processing efficiency of thermal imaging videos, and reduces the operational threshold and equipment usage costs.

[0068] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of a non-contact sarcopenia risk assessment system based on multimodal thermal imaging according to an embodiment of this application. This application proposes a non-contact sarcopenia risk assessment system based on multimodal thermal imaging, comprising:

[0069] The data acquisition module is used to acquire thermal imaging videos and RGB videos of the subject at multiple motion phases;

[0070] The region tracking module is used to perform key point detection and region of interest localization on thermal imaging video and RGB video to obtain the first quadriceps muscle region of interest data. Then, the first quadriceps muscle region of interest data is subjected to quality inspection and motion compensation to obtain the second quadriceps muscle region of interest data.

[0071] The feature extraction module is used to extract temporal features from the region of interest data of the second quadriceps muscle to obtain a temporal feature vector, and to calculate the reliability of the region of interest data of the second quadriceps muscle to obtain a reliability index.

[0072] The risk assessment module is used to input time-series feature vectors into multiple pre-trained sarcopenia risk prediction models to obtain sarcopenia risk prediction values. Based on the sarcopenia risk prediction values ​​and reliability indicators, the sarcopenia risk assessment results are obtained.

[0073] The time-series feature vector includes the baseline temperature mean, maximum temperature rise, recovery half-life, metabolic recovery index, and thermo-metabolic efficiency.

[0074] It should be noted that the embodiments of this application adopt an integrated design, including four functional modules: data acquisition, dynamic area tracking, thermodynamic feature extraction, and AI risk assessment, to construct a complete non-contact sarcopenia risk assessment process. Among them, a non-contact dynamic monitoring method is adopted, using infrared thermal imaging technology to capture the temperature changes in standardized movements such as sitting leg extension, thus avoiding the risks of radiation and contact. This method can not only extract dynamic indicators such as temperature rise slope and thermal recovery half-life to evaluate neuromuscular coordination efficiency, but also achieve full-process automation, greatly improving the processing efficiency of thermal imaging videos, reducing the operation threshold and equipment usage cost.

[0075] Specifically, the entire process strictly follows the standardized movement protocol. Under the guidance of a metronome, the subject completes the sitting leg extension movement (starting posture: sitting on a chair with a backrest, the torso supported by the backrest, and the hip, knee, and ankle joints all at 90°; both hands hold the sides of the chair to maintain balance). The knee joint is slowly extended within 0-2 seconds until the angle between the thigh and the calf reaches 170°, and then the knee is controlled to bend back to the initial posture within 2-4 seconds. The infrared thermal imager continuously captures the thermal video images during the baseline period, movement period, and recovery period to ensure that the temperature change process caused by muscle activity is captured.

[0076] As Figure 2 shown is the schematic diagram of the three-stage standardized data acquisition process. In the continuous acquisition mode, the subject completes a full detection process, and the system continuously acquires the whole-process data and automatically identifies and splits the three stages subsequently.

[0077] Stage 1 is pre-movement baseline acquisition. The subject maintains a sitting posture to acquire the baseline thermal image data of the quadriceps femoris. The acquisition duration can be set to any value within the range of 10 to 20 seconds according to the equipment preheating and infrared thermal imaging stabilization settings. In the embodiments of this application, the acquisition duration is set to 15 seconds to establish a stable baseline. The automatic recognition criterion is that when the temperature variance satisfies std(T) < k×σ_sensor (k = 1.0 - 1.5, σ_sensor is estimated by a blackbody or equipment self-check), and there is no obvious movement trajectory, it is confirmed as the baseline period. If there is no external blackbody, σ_sensor is estimated by the built-in shutter of the equipment or the static background reference.

[0078] Phase 2 is the standardized exercise period. Subjects complete slow seated leg extension movements under the guidance of a metronome. The data collection duration can be set to any value within the range of 30 to 60 seconds, depending on actual needs. In this embodiment, the data collection duration is set to 40 seconds. The preferred exercise parameters are 8 leg extension movements, each lasting 4 seconds (2 seconds of extension + 2 seconds of retraction), which can be adjusted within the range of 6-12 repetitions according to individual differences to ensure the rhythm and repeatability of the movements. Automatic identification of the exercise period is based on the detection of obvious periodic movement patterns and temperature fluctuations. Adjustable movement amplitude is set for elderly individuals, prioritizing comfortable knee extension and avoiding excessive force.

[0079] Phase 3 is the post-exercise recovery period. Subjects sit still to collect data on the muscle thermometabolic recovery process. The data collection duration can be set to any value within the range of 90 to 180 seconds, depending on the complete muscle recovery time. In this embodiment, the data collection duration is set to 120 seconds to observe the complete recovery curve. It should be noted that clinical studies support the finding that the muscle recovery half-life in healthy elderly individuals is 40-70 seconds, and in patients with sarcopenia, it is 70-110 seconds. Therefore, the 120-second detection window set in this embodiment can adequately observe the completeness of recovery. The automatic identification of the start of the recovery period is based on the cessation of exercise and an exponential or stretching-exponential decrease in temperature.

[0080] Simultaneous acquisition of thermal imaging video is performed using dual-spectral synchronous acquisition. Thermal imaging video is acquired via an infrared thermal imager (wavelength range 8-14 micrometers, frame rate 30-60Hz, resolution 640×480 or higher), while RGB video is acquired via a visible light camera. Synchronization is achieved using hardware triggering or a unified timestamp, with a synchronization error ≤ ±20ms. For equipment calibration and temperature reference, a portable blackbody is selected as the temperature reference for absolute temperature calibration and σ_sensor estimation. In the absence of an external blackbody, a built-in shutter reference or a stable background area can be used for relative calibration. Environmental recording and temperature drift compensation include recording ambient temperature / humidity, airflow, and surface emissivity settings, and performing environmental temperature drift compensation to reduce the systematic influence of external factors on the curve. The total acquisition time is set to 175 seconds, but can be adaptively adjusted within the range of 150 to 240 seconds based on individual needs and stability. The output data consists of thermal video sequences H_pre(t), H_exe(t), and H_rec(t) corresponding to three motion stages, along with the corresponding RGB video sequences.

[0081] As a further optional implementation, the exercise phase includes a baseline period, an exercise period, and a recovery period, and the area tracking module includes:

[0082] The key point detection module is used to calibrate RGB video and thermal imaging video, and then perform key point detection on the thermal imaging video corresponding to the movement period based on the calibrated RGB video to obtain the key points of the lower limbs.

[0083] The boundary segmentation module is used to segment the quadriceps femoris boundary in thermal imaging videos based on key points of the lower limbs to obtain the optimal region of interest.

[0084] The boundary alignment module is used to align the optimal region of interest with the thermal imaging videos corresponding to the baseline and recovery periods to obtain the region of interest data of the first quadriceps muscle for each movement phase.

[0085] Specifically, this application employs a combined geometric and thermal calibration scheme, using a dedicated thermal imaging calibration board or a portable blackbody array to complete the intrinsic and extrinsic parameter calibration of RGB and thermal images. The coarse localization stage processes the three-stage RGB video sequence, focusing on utilizing the H_exe(t) data from the movement phase and using OpenPose to detect lower limb key points. In the precise segmentation stage, a lightweight U-Net is used for quadriceps boundary segmentation within the thermal imaging domain. The aforementioned acquired RGB video is only used for initial ROI cues and key point references to reduce cross-modal errors. The optimal region of interest (ROI) boundary determined during the movement phase is propagated to the baseline phase H_pre(t) and the recovery phase H_rec(t). Based on the automatically identified three-stage boundaries, the ROI within each movement phase is fine-tuned and optimized. The output first quadriceps ROI data consists of a three-stage quadriceps ROI coordinate sequence {ROI_pre(t), ROI_exe(t), ROI_rec(t)} and temporal boundary markers.

[0086] As a further optional implementation, the exercise phase includes a baseline period, an exercise period, and a recovery period, and the area tracking module includes:

[0087] The quality inspection module is used to detect abnormal frames in the region of interest data of the first quadriceps muscle corresponding to each movement stage, and to perform spatial alignment verification on the region of interest data of the first quadriceps muscle corresponding to each movement stage to identify abnormal situations and then repair them.

[0088] The motion compensation module is used to perform thermal stability correction on the repaired region of interest (ROI) data of the first quadriceps femoris during the baseline phase, global displacement compensation on the repaired ROI data of the first quadriceps femoris during the exercise phase, and static compensation on the repaired ROI data of the first quadriceps femoris during the recovery phase, so as to obtain the ROI data of the second quadriceps femoris for each exercise phase.

[0089] Specifically, a three-stage temporal consistency check was first performed on the quadriceps muscle region of interest (ROI) data, including intra-stage continuity and inter-stage consistency. Intra-stage continuity was determined by detecting abnormal frames, while inter-stage consistency verified the spatial correspondence of the ROIs in the three stages to ensure that the same muscle region was analyzed. For abnormal cases, temporal interpolation or frame smoothing was used for repair. Subsequently, thermal imaging and RGB pixel-level alignment verification were performed to ensure accurate mapping of the ROI boundaries on the thermal image. Finally, segmentation and registration were completed in the thermal imaging domain.

[0090] The phased motion compensation strategy is as follows: during the baseline period, there is almost no motion, and the focus is on thermal stability correction; during the motion period, ECC affine or rigid registration is used, or global displacement compensation is performed using sparse optical flow (Lucas-Kanade); during the recovery period, static compensation is used, and rigid or translational models can be selected to ensure ROI stability. Data quality assessment measures temperature data integrity, dynamic range, and ROI stability in stages, and the ROI stability score is included in the reliability index R_quality. The output quadriceps region of interest data is the quality-verified three-stage quadriceps region of interest thermal video sequence {H'_pre(t), H'_exe(t), H'_rec(t)} and corresponding quality labels.

[0091] As a further optional implementation, the feature extraction module includes:

[0092] The statistical temperature calculation module is used to extract the temperature values ​​of all pixels in the region of interest data of the second quadriceps muscle, construct a temperature matrix, and then calculate the statistical temperature of the region of interest based on the temperature matrix.

[0093] The baseline feature extraction module is used to determine the first sampling window. Within the first sampling window, the average baseline temperature is calculated based on the statistical temperature of the region of interest.

[0094] The motion period feature extraction module is used to determine the second sampling window. Within the second sampling window, the peak search is performed on the statistical temperature of the region of interest, and then the maximum temperature rise is calculated based on the peak search result and the baseline temperature mean.

[0095] The recovery period feature extraction module is used to determine the third sampling window. Within the third sampling window, the recovery half-life, recovery integrity, and recovery slope are calculated based on the statistical temperature of the region of interest, the mean baseline temperature, the peak search result, and the maximum temperature rise.

[0096] The feature index calculation module is used to calculate the metabolic recovery index based on the recovery integrity and the thermo-metabolic efficiency based on the maximum temperature rise and the recovery slope.

[0097] Specifically, the three-stage temporal analysis algorithm performs joint analysis on the region of interest data {H'_pre(t), H'_exe(t), H'_rec(t)} of the second quadriceps muscle, focusing on the post-exercise recovery process.

[0098] First, temperature time-series data preprocessing and ROI temperature calculation are performed. For each frame of thermal image H'(t), the temperature values ​​of all pixels are extracted within the located quadriceps muscle region of interest, constructing a pixel-level temperature matrix T_ROI(t) = {T_pixel(i,j,t)|(i,j)∈ROI}, t∈[0,175s]. The statistical temperature T_mean(t) of the region of interest is then calculated as the representative temperature at that moment using the following formula:

[0099] T_mean(t) = mean(T_ROI(t));

[0100] After obtaining the statistical temperature T_mean(t) of the region of interest, extreme pixels are removed according to the 3σ principle to prevent thermal reflection interference. The time series data is then smoothed to obtain the smoothed statistical temperature T_smooth(t) of the region of interest. Savitzky-Golay filtering (window length 0.5-1.0 seconds, corresponding to 15-60 frames; polynomial order 2-3) is used to balance smoothness and peak / slope fidelity, ensuring the accuracy of the timestamps of the 30-60Hz sampling rate data (accuracy ±0.1 seconds). Linear interpolation is used to complete the time series data for occasional frame drops.

[0101] Next, baseline features are extracted from the first 15 seconds of collected data (i.e., baseline period data). The mean baseline temperature T_baseline is calculated within a window of t∈[5s,15s] (i.e., the first sampling window) to avoid the initial unstable period of the first 5 seconds. The mean baseline temperature T_baseline is calculated using the following formula:

[0102] T_baseline=(1 / N)∑[t=5s to 15s]T_smooth(t);

[0103] Where N is the number of sampling points, and the temperature unit is degrees Celsius. Baseline temperature stability (std_baseline) is defined as:

[0104]

[0105] The baseline is considered stable when the baseline temperature stability std_baseline < 0.1℃; otherwise, it is marked as low-quality data. Linear regression is used to detect temperature drift during the baseline period, with a slope of < 0.01℃ / s.

[0106] Then, motion-period feature extraction is performed on the collected data from 15 to 55 seconds (i.e., motion-period data). The maximum temperature rise ΔT_max is calculated within the window t∈[15s,55s] (i.e., the second sampling window). First, peak search is performed, defining T_peak=max{T_smooth(t)|t∈[15s,55s]}, and the maximum temperature rise ΔT_max is calculated using the following formula, and the peak time t_peak is recorded:

[0107] ΔT_max=T_peak-T_baseline;

[0108] t_peak=argmax{T_smooth(t)|t∈[15s,55s]};

[0109] The rising phase, from baseline stabilization to reaching 90% peak value, is assessed using a least-squares linear fit, T(t) = a·t + b, to extract the temperature rise rate dT / dt_up = a (unit: °C / s). The motion response time is defined as T_response = t_peak - 15 s, reflecting the rate of muscle metabolic activation. To improve the robustness of peak detection, a Hampel filter is applied to the statistical temperature T_smooth of the region of interest, and peak values ​​are searched within neighborhood constraints to suppress misjudgments of t_peak caused by isolated spikes.

[0110] Furthermore, recovery period features are extracted from the collected data (55-175 seconds, i.e., recovery period data). The recovery half-life t_1 / 2, recovery integrity dT / dt_down, and recovery slope R_complete are calculated within a window (i.e., the third sampling window) of t∈[55s,175s]. The recovery half-life t_1 / 2 is based on the target temperature T_half=T_baseline+0.5×(T_peak-T_baseline), and is searched according to t_1 / 2=min{t|T_smooth(t)≤T_half,t>t_peak}. If no exact matching point is found, a piecewise linear interpolation method with monotonicity constraints is used to improve the time accuracy to 0.1 seconds. If the half-life is not reached within 120 seconds, t_1 / 2 is set to 120s and an anomaly is marked.

[0111] The recovery slope dT / dt_down is extracted from the initial descent segment (t_peak to t_peak+30s) using a linear fit T(t) = c·t+d, with units of ℃ / s (negative values ​​indicate a downward trend); the recovery integrity R_complete is calculated with T_120s = T_smooth(175s) as the endpoint temperature, defined as R_complete = (T_peak-T_120s) / (T_peak-T_baseline)×100%, and restricted to R_complete ∈ [0%, 100%], with values ​​exceeding 100% being truncated to 100%.

[0112] Furthermore, the recovery curve decay model fitting used single-exponential, double-exponential, and stretching exponential models as candidate models, respectively:

[0113] T(t)=T_baseline+(T_peak-T_baseline)×exp(-(t-t_peak) / τ);

[0114] T(t)=T_baseline+A·exp(-(t-t_peak) / τ_1)+B·exp(-(t-t_peak) / τ_2);

[0115] T(t)=T_baseline+(T_peak-T_baseline)×exp(-((t-t_peak) / τ)^β);

[0116] The optimal model is automatically selected based on AIC / BIC, and the selected model type and parameters are recorded. Parameter estimation uses nonlinear least squares method, and the fitting window is t∈[t_peak,175s]. The fitting quality is measured by the coefficient of determination R0. 2 Measurement, preferred R 2 The value is ≥0.85, and residual diagnosis is output. The recovery stability assessment calculates the residual sequence between the actual temperature and the fitted curve through residual analysis, and uses the residual standard deviation σ_residual < 0.2℃ as the stability criterion. At the same time, first-order difference is used to detect abnormal fluctuation points.

[0117] Finally, comprehensive characteristic indicators are calculated. The metabolic recovery index MRI is calculated using the following formula:

[0118] MRI=t_1 / 2_norm×(2-R_complete_weight);

[0119] Wherein, t_1 / 2_norm = t_1 / 2 / 60s (based on 60 seconds), and R_complete_weight = R_complete / 100%. The lower the metabolic recovery index on MRI, the better the recovery ability; healthy elderly people have an MRI index <0.8, while patients with sarcopenia have an MRI index >1.2.

[0120] Thermometabolic efficiency (HME) is used to comprehensively reflect the capacity for heat production and dissipation, and it is calculated using the following formula:

[0121] HME = ΔT_max / |dT / dt_down| (unit: °C·s);

[0122] ΔT_max reflects the muscle's heat production capacity, while |dT / dt_down| reflects the muscle's heat dissipation capacity. A lower thermometabolic efficiency (HME) indicates higher heat dissipation efficiency and healthier muscle metabolism.

[0123] As a further optional implementation, the feature extraction module includes:

[0124] The data integrity assessment module is used to determine the missing proportion and the stability of the region of interest, and then calculates the data integrity score based on the missing proportion and the stability of the region of interest.

[0125] The signal-to-noise ratio (SNR) evaluation module is used to determine the baseline temperature stability, and then calculate the SNR score based on the maximum temperature rise and the baseline temperature stability.

[0126] The fit quality assessment module is used to determine the goodness of fit of the recovery curve, and then calculates the fit quality score based on the goodness of fit of the recovery curve and the baseline temperature stability.

[0127] The physiological rationality assessment module is used to assess the range of baseline temperature mean, maximum temperature rise, and recovery half-life, and then calculate the physiological rationality score based on the range assessment results.

[0128] The reliability index calculation module is used to determine the weight values. Based on the weight values, the data integrity score, signal-to-noise ratio score, fitting quality score, and physiological rationality score are weighted and summed to obtain the reliability index.

[0129] Specifically, the data reliability index R_quality is based on the data integrity score (completeness) and the signal-to-noise ratio score.

[0130] The scores are calculated as SNR_score, fitting_score, and physiological rationality score.

[0131] For data integrity assessment, a weight value of 0.3 is determined. First, the missing ratio (missing_ratio) is determined as the number of missing frames divided by the total number of frames. The region of interest (ROI) stability is set to min(1, max(0, 1-α·δ_area-β·δ_shift)), where δ_area is the area fluctuation rate within a stage, and δ_shift is the relative threshold exceeding the center displacement. The displacement is normalized according to the image diagonal or ROI diameter. With α = 0.5 and β = 0.5, data integrity (completeness) is calculated using the following formula:

[0132] completeness=(1-missing_ratio)×roi_stability;

[0133] For signal-to-noise ratio (SNR) evaluation, a weight value of 0.25 is determined. Let signal = maximum temperature rise ΔT_max and noise = baseline temperature stability std_baseline. The SNR is calculated as signal / noise, and the SNR score is obtained using the following formula:

[0134] SNR_score=min(SNR / 10,1.0);

[0135] For the fitting quality assessment, a weight value of 0.25 was determined. First, the goodness of fit Rfit of the overall recovery curve was calculated. 2 Given that _recovery and baseline stability = exp(-std_baseline / 0.1), the fitting quality score _fitting_score is calculated using the following formula:

[0136] fitting_score = 0.6 × R 2 _recovery + 0.4 × stability;

[0137] For the physiological rationality assessment, the corresponding weight value is determined to be 0.2. Examining the ranges of maximum temperature rise ΔT_max ∈ [0.5℃, 5.0℃], recovery half-life t_1 / 2 ∈ [10s, 120s], and baseline temperature mean T_baseline ∈ [28℃, 37℃], the physiological baseline temperature mean is calculated using the following formula:

[0138] physiological = (number of indicators within the range) / (total number of indicators);

[0139] The final reliability index R_quality is calculated using the following formula:

[0140] R_quality=0.3×completeness+0.25×SNR_score+0.25×fitting_score+0.2×physiological;

[0141] The quality level is defined as follows: reliability index R_quality > 0.8 is high quality, 0.6 < reliability index R_quality < 0.8 is medium quality, and reliability index R_quality < 0.6 is low quality. Low quality data is automatically marked and its weight is reduced or excluded in the final analysis.

[0142] After obtaining the various time-series feature vectors and reliability metrics, the time-series feature vectors are constructed.

[0143] F = [T_baseline, ΔT_max, t_1 / 2, MRI, HME, τ] (F ∈ R^6, component precision is retained to 2 decimal places), and Z-score normalization is performed based on the statistical characteristics of the training data. The output includes the feature vector F, the reliability index R_quality, and auxiliary metadata (fitting parameters, intermediate calculation results, anomaly markers), while also recording the extraction time and data time range for each feature.

[0144] As an optional implementation, the non-contact sarcopenia risk assessment system based on multimodal thermal imaging further includes a parameter correction module, which includes:

[0145] The parameter acquisition module is used to acquire the subject's age, gender, body mass index, and medical history, and to collect environmental temperature and humidity data.

[0146] The correction coefficient determination module is used to determine the first correction coefficient based on age and gender, the second correction coefficient based on body mass index, the third correction coefficient based on medical history, and the fourth correction coefficient based on environmental temperature and humidity data.

[0147] The feature vector correction module is used to correct the time series feature vector based on the first correction coefficient, the second correction coefficient, the third correction coefficient, and the fourth correction coefficient to obtain the corrected time series feature vector.

[0148] Specifically, such as Figure 3The diagram illustrates the multi-level correction process. The multi-level correction model includes physiological baseline correction (establishing a basal metabolic temperature model based on age and sex), body parameter correction (considering the nonlinear effects of body mass index (BMI) and body surface area on heat dissipation), environmental adaptive correction (real-time environmental temperature and humidity compensation algorithm), and pathological state prior (Bayesian correction based on past medical history and physical examination indicators). Regarding the correction algorithm, the correction of the baseline mean temperature T_baseline is T'_baseline = T_baseline + α1 × age_offset + β1 × BMI_correction; dynamic features are corrected using piecewise linear or polynomial models, and the interaction effects of age × sex and BMI × baseline temperature are modeled. Methodologically, hierarchical Bayesian or linear mixed-effects models are preferred for individualized / group hierarchical decomposition, with Gaussian process regression serving as a nonlinear supplement; correction parameters are fitted during the training phase and frozen during the inference phase to avoid data leakage and overfitting. A 95% confidence interval is output for the corrected key features. The final output consists of the corrected feature vector F', the correction confidence C_correction, and the individual's physiological profile.

[0149] As an optional implementation, the sarcopenia risk prediction model includes a first risk prediction model, a second risk prediction model, and a third risk prediction model. The non-contact sarcopenia risk assessment system based on multimodal thermal imaging also includes a model training module, which includes:

[0150] The training data acquisition module is used to acquire the training dataset, which includes several training samples and the corresponding labeled data for each training sample.

[0151] The training data processing module is used to extract features from the training samples to obtain sample feature vectors, and to collect the age, gender, body mass index, and medical history of the training samples to obtain extended features of the samples.

[0152] The first model training module is used to input the sample feature vector and sample extended features into the preset gradient boosting decision tree model for training, and to optimize the parameters of the trained gradient boosting decision tree model based on the labeled data to obtain the first risk prediction model.

[0153] The second model training module is used to input the sample feature vector and sample extended features into a preset random forest model for training, and to optimize the parameters of the trained random forest model based on the labeled data to obtain the second risk prediction model.

[0154] The third model training module is used to input the sample feature vectors and sample extended features into a preset support vector machine model for training, and to optimize the parameters of the trained support vector machine model based on the labeled data to obtain the third risk prediction model.

[0155] Specifically, 300-500 labeled data were collected, including 150-200 confirmed sarcopenia patients, 150-200 healthy controls, and 50-100 borderline cases, forming a data matrix X_train=[F_1,F_2,...,F_n]^T∈R^{n×6} and a label vector Y_train=[y_1,y_2,...,y_n]^T(y_i∈{0=healthy,1=borderline,2=sarcopenia}). At the same time, physiological parameters such as age, gender, and body mass index (BMI) were collected as sample expansion features.

[0156] Next, all 300-500 samples underwent a complete processing flow through the aforementioned data acquisition, region tracking, and feature extraction modules, with only high- to medium-quality samples (R_quality > 0.6) retained for training. A standard feature vector F_i = [T_baseline_i, ΔT_max_i, t_1 / 2_i, MRI_i, HME_i, τ_i] was generated for each sample, and a gold standard clinical label was established based on DXA muscle mass measurement, grip strength test, and walking speed. Furthermore, the Z-score normalization parameters μ and σ (X_norm = (X-μ) / σ) were calculated, outliers were handled using the 3σ principle, and the optimal feature subset was selected through analysis of variance and mutual information.

[0157] During model training, recovery half-life t_1 / 2, recovery completeness R_complete, and metabolic recovery index MRI are used as core features, supplemented by baseline mean temperature T_baseline, maximum temperature rise ΔT_max, thermometabolic efficiency HME, age, and body mass index BMI. Random forest is used to calculate feature importance to retain key features, and interactive features such as age×MRI and BMI×t_1 / 2 are constructed. In this embodiment, a weighted voting ensemble strategy of three models is adopted, with weights determined by validation set performance. The gradient boosting decision tree (XGBoost / LightGBM) model is the main one, suitable for small sample medical data. The parameter optimization range is learning_rate∈[0.01,0.1], max_depth∈[3,8], and n_estimators∈[100,500]. Overfitting is suppressed by L1 / L2 regularization and early_stopping. The alternative models include a random forest (n_trees=100-300) module and a support vector machine (RBF kernel, C∈[0.1,10]) module.

[0158] During training and optimization, class weight and threshold tuning are prioritized to address class imbalance. SMOTE is only used when rigorously validating the introduction of pseudo-models. Bayesian optimization (TPE) is employed to search for optimal hyperparameters, targeting the F1-score. Five-fold hierarchical cross-validation is performed, and an independent external validation set is set for generalization evaluation. Probability calibration is performed on the validation set using Platt or Isotonic methods to evaluate and minimize ECE / ACE. The rejection mechanism outputs "retest / further investigation required" when the maximum predicted probability is below a threshold (e.g., 0.6-0.7) or R_quality < 0.6. Model evaluation and selection are based on metrics such as accuracy, precision, recall, F1-score, and AUC-ROC, combined with a confusion matrix to focus on sarcopenia recall. Consistency analysis with the gold standard diagnosis is also performed (Kappa coefficient > 0.7) to obtain the trained first-risk prediction model, second-risk prediction model, and third-risk prediction model.

[0159] As an optional implementation, the sarcopenia risk prediction model includes a first risk prediction model, a second risk prediction model, and a third risk prediction model, and the risk assessment module includes:

[0160] The first model prediction module is used to input the corrected time-series feature vector into the first risk prediction model and output the first sarcopenia probability value.

[0161] The second model prediction module is used to input the corrected time-series feature vector into the second risk prediction model and output the second sarcopenia probability value.

[0162] The third model prediction module is used to input the corrected time series feature vector into the third risk prediction model and output the third sarcopenia probability value.

[0163] The risk prediction value determination module is used to perform a weighted summation of the first, second, and third sarcopenia probability values, and then determine the maximum value in the weighted summation result as the sarcopenia risk prediction value.

[0164] Specifically, the corrected temporal feature vector F ' Input the trained first, second, and third risk prediction models, and issue low-confidence warnings for low-quality samples with a reliability index R_quality < 0.6. The three trained models respectively assess F... 'Predictions were made to obtain the XGBoost probability P_xgb = [p_0, p_1, p_2] (i.e., the first sarcopenia probability value), the Random Forest probability P_rf = [p_0, p_1, p_2] (i.e., the second sarcopenia probability value), and the SVM probability P_svm = [p_0, p_1, p_2] (i.e., the third sarcopenia probability value). Then, a weighted fusion was performed, P_final = w_xgb × P_xgb + w_rf × P_rf + w_svm × P_svm, and predicted_class = argmax(P_final) was used as the final sarcopenia risk prediction value.

[0165] As an optional implementation, the risk assessment module includes:

[0166] The confidence calculation module is used to calculate the confidence level based on the sarcopenia risk prediction value and reliability index;

[0167] The risk assessment output module is used to output sarcopenia risk assessment results based on confidence level and sarcopenia risk prediction value.

[0168] Specifically, the confidence calculation comprehensively considers the consistency of the model and the probability certainty. Among them, max(R_quality)>0.8 is considered as high confidence, and the influence of data quality is introduced: confidence_final=max(P_final)×R_quality_test. When max(confidence_final)<0.6, further examination is recommended. The final output of the sarcopenia risk assessment results includes the sarcopenia risk probability P_final[2]×100%, the result confidence_final×100%, and the key indicator analysis, showing the first three abnormal characteristics and their clinical meaning. For example, the recovery half-life t_1 / 2=85 seconds (normal <50 seconds) indicates a decrease in muscle recovery ability, and the metabolic recovery index MRI=1.5 (normal <1.0) indicates abnormal metabolic recovery function. The risk of sarcopenia is classified into three categories: low risk (predicted_class=0, sarcopenia probability <20%, routine follow-up recommended), medium risk (predicted_class=1, sarcopenia probability 20%-60%, follow-up recommended every 3-6 months), and high risk (predicted_class=2, sarcopenia probability >60%, immediate medical attention recommended for diagnosis).

[0169] The foregoing has described the non-contact sarcopenia risk assessment system based on multimodal thermal imaging according to embodiments of this application. It can be appreciated that the embodiments of this application, including a data acquisition module, a region tracking module, a feature extraction module, and a risk assessment module, have the following advantages:

[0170] First, it has the advantage of non-contact safety testing, which can effectively avoid the radiation risks (such as DEXA) and inconveniences caused by contact operation (such as BIA) in existing technologies. It is suitable for high-frequency community screening, protecting residents' health while improving the accessibility of testing.

[0171] Second, by quantifying neuromuscular synergy efficiency through key indicators such as baseline temperature mean, maximum temperature rise, recovery half-life, metabolic recovery index, and thermometabolic efficiency, the limitations of traditional static muscle mass detection are overcome. Furthermore, three neural network models are used to achieve dynamic functional assessment of sarcopenia. The probability values ​​output by the three neural network models are weighted and summed to obtain the final sarcopenia risk prediction value, which can provide a more accurate sarcopenia risk prediction value and greatly improve the processing efficiency of thermal imaging video.

[0172] Third, it requires no professional personnel to operate, making it easy to use and greatly lowering the barrier to entry. It also has wide applicability, especially suitable for various special populations. Because the test can be completed with only simple movements, it is friendly to individuals with severe muscle weakness and obesity, solving the problem of inaccurate detection by body composition analyzers in obese patients.

[0173] Fourth, the cost of terminal equipment is significantly lower than that of professional equipment such as DEXA. It not only supports the rapid deployment of new equipment, but can also be used to upgrade and transform existing medical equipment, thereby improving the overall cost-effectiveness and application flexibility.

[0174] Reference Figure 4 This application also provides a non-contact sarcopenia risk assessment method based on multimodal thermal imaging, including the following steps S101 to S104:

[0175] Step S101: Acquire thermal imaging videos and RGB videos of the subject during multiple motion phases;

[0176] Step S102: Perform key point detection and region of interest localization on the thermal imaging video and RGB video to obtain the first quadriceps muscle region of interest data. Then, perform quality checks and motion compensation on the first quadriceps muscle region of interest data to obtain the second quadriceps muscle region of interest data.

[0177] Step S103: Extract temporal features from the region of interest data of the second quadriceps femoris to obtain a temporal feature vector, and calculate the reliability of the region of interest data of the second quadriceps femoris to obtain a reliability index;

[0178] Step S104: Input the time-series feature vector into multiple pre-trained sarcopenia risk prediction models to obtain sarcopenia risk prediction values. Based on the sarcopenia risk prediction values ​​and reliability indicators, obtain the sarcopenia risk assessment results.

[0179] The time-series feature vector includes the baseline temperature mean, maximum temperature rise, recovery half-life, metabolic recovery index, and thermo-metabolic efficiency.

[0180] It is understood that the content of the above system embodiments is applicable to this method embodiment. The specific functions implemented in this method embodiment are the same as those in the above system embodiments, and the beneficial effects achieved are also the same as those achieved in the above system embodiments.

[0181] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0182] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0183] Please see Figure 5 , Figure 5 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0184] The processor 1001 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0185] The memory 1002 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1002 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1002 and is called and executed by the processor 1001 using the methods described in the embodiments of this application.

[0186] Input / output interface 1003 is used to implement information input and output;

[0187] The communication interface 1004 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0188] Bus 1005 transmits information between various components of the device (e.g., processor 1001, memory 1002, input / output interface 1003, and communication interface 1004);

[0189] The processor 1001, memory 1002, input / output interface 1003 and communication interface 1004 are connected to each other within the device via bus 1005.

[0190] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0191] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0192] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0193] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0194] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0195] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the aforementioned blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0196] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the aforementioned functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the realm of conventional skill for an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0197] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a 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, server, or 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, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0198] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0199] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0200] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0201] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A non-contact sarcopenia risk assessment system based on multimodal thermal imaging, characterized in that, include: The data acquisition module is used to acquire thermal imaging videos and RGB videos of the subject at multiple motion phases; The region tracking module is used to perform key point detection and region of interest localization on the thermal imaging video and the RGB video to obtain the first quadriceps muscle region of interest data, and then perform quality checks and motion compensation on the first quadriceps muscle region of interest data to obtain the second quadriceps muscle region of interest data. The feature extraction module is used to extract temporal features from the second quadriceps muscle region of interest data to obtain a temporal feature vector, and to calculate the reliability of the second quadriceps muscle region of interest data to obtain a reliability index. The risk assessment module is used to input the time-series feature vector into multiple pre-trained sarcopenia risk prediction models to obtain sarcopenia risk prediction values, and to obtain sarcopenia risk assessment results based on the sarcopenia risk prediction values ​​and the reliability index. The time-series feature vector includes the baseline temperature mean, maximum temperature rise, recovery half-life, metabolic recovery index, and thermal metabolic efficiency.

2. The system according to claim 1, characterized in that, The motion phase includes a baseline period, a motion period, and a recovery period. The region tracking module includes: The key point detection module is used to calibrate the RGB video and the thermal imaging video, and then perform key point detection on the thermal imaging video corresponding to the movement period based on the calibrated RGB video to obtain the key points of the lower limbs. The boundary segmentation module is used to segment the quadriceps femoris boundary in the thermal imaging video based on the key points of the lower limbs to obtain the optimal region of interest. The boundary alignment module is used to align the optimal region of interest with the thermal imaging videos corresponding to the baseline period and the recovery period to obtain the first quadriceps region of interest data corresponding to each of the motion stages.

3. The system according to claim 1, characterized in that, The motion phase includes a baseline period, a motion period, and a recovery period. The region tracking module includes: The quality inspection module is used to detect abnormal frames in the first quadriceps muscle region of interest data corresponding to each of the movement stages, and to perform spatial alignment verification on the first quadriceps muscle region of interest data corresponding to each of the movement stages to obtain abnormal situations, and then repair the abnormal situations. The motion compensation module is used to perform thermal stability correction on the repaired region of interest data of the first quadriceps femoris corresponding to the baseline period, perform global displacement compensation on the repaired region of interest data of the first quadriceps femoris corresponding to the exercise period, and perform static compensation on the repaired region of interest data of the first quadriceps femoris corresponding to the recovery period, so as to obtain the region of interest data of the second quadriceps femoris corresponding to each exercise stage.

4. The system according to claim 1, characterized in that, The feature extraction module includes: The statistical temperature calculation module is used to extract all pixel temperature values ​​in the second quadriceps muscle region of interest data, construct a temperature matrix, and then calculate the statistical temperature of the region of interest based on the temperature matrix. The baseline feature extraction module is used to determine a first sampling window, and within the first sampling window, calculate the mean baseline temperature based on the statistical temperature of the region of interest. The motion period feature extraction module is used to determine the second sampling window. Within the second sampling window, the peak search is performed on the statistical temperature of the region of interest, and then the maximum temperature rise is calculated based on the peak search result and the baseline temperature mean. The recovery period feature extraction module is used to determine the third sampling window. Within the third sampling window, the recovery half-life, recovery integrity, and recovery slope are calculated based on the statistical temperature of the region of interest, the mean baseline temperature, the peak search result, and the maximum temperature rise. The feature index calculation module is used to calculate the metabolic recovery index based on the recovery integrity, and to calculate the thermal metabolic efficiency based on the maximum temperature rise and the recovery slope.

5. The system according to claim 1, characterized in that, The feature extraction module includes: The data integrity assessment module is used to determine the missing proportion and the stability of the region of interest, and then calculate the data integrity score based on the missing proportion and the stability of the region of interest. The signal-to-noise ratio (SNR) evaluation module is used to determine the baseline temperature stability and then calculate the SNR score based on the maximum temperature rise and the baseline temperature stability. The fitting quality assessment module is used to determine the goodness of fit of the recovery curve, and then calculate the fitting quality score based on the goodness of fit of the recovery curve and the baseline temperature stability. The physiological rationality assessment module is used to assess the range of the baseline temperature mean, the maximum temperature rise, and the recovery half-life, and then calculate the physiological rationality score based on the range assessment results. The reliability index calculation module is used to determine the weight values, and based on the weight values, to perform a weighted summation of the data integrity score, the signal-to-noise ratio score, the fitting quality score, and the physiological rationality score to obtain the reliability index.

6. The system according to claim 1, characterized in that, The non-contact sarcopenia risk assessment system based on multimodal thermal imaging also includes a parameter correction module, which comprises: The parameter acquisition module is used to acquire the subject's age, gender, body mass index, and medical history, and to collect environmental temperature and humidity data. The correction coefficient determination module is used to determine a first correction coefficient based on the age and the gender, a second correction coefficient based on the body mass index, a third correction coefficient based on the medical history, and a fourth correction coefficient based on the environmental temperature and humidity data. The feature vector correction module is used to correct the time series feature vector according to the first correction coefficient, the second correction coefficient, the third correction coefficient and the fourth correction coefficient to obtain the corrected time series feature vector.

7. The system according to claim 1, characterized in that, The sarcopenia risk prediction model includes a first risk prediction model, a second risk prediction model, and a third risk prediction model. The non-contact sarcopenia risk assessment system based on multimodal thermal imaging also includes a model training module, which includes: The training data acquisition module is used to acquire a training dataset, which includes a number of training samples and labeled data corresponding to each training sample. The training data processing module is used to extract features from the training samples to obtain sample feature vectors, and to collect the age, gender, body mass index, and medical history of the training samples to obtain extended features of the samples. The first model training module is used to input the sample feature vector and the sample extended features into a preset gradient boosting decision tree model for training, and to optimize the parameters of the trained gradient boosting decision tree model according to the labeled data to obtain the first risk prediction model. The second model training module is used to input the sample feature vector and the sample extended features into a preset random forest model for training, and to optimize the parameters of the trained random forest model based on the labeled data to obtain the second risk prediction model. The third model training module is used to input the sample feature vector and the sample extended features into a preset support vector machine model for training, and to optimize the parameters of the trained support vector machine model based on the labeled data to obtain the third risk prediction model.

8. The system according to claim 6, characterized in that, The sarcopenia risk prediction model includes a first risk prediction model, a second risk prediction model, and a third risk prediction model. The risk assessment module includes: The first model prediction module is used to input the corrected time-series feature vector into the first risk prediction model and output the first sarcopenia probability value. The second model prediction module is used to input the corrected time-series feature vector into the second risk prediction model and output the second sarcopenia probability value. The third model prediction module is used to input the corrected time-series feature vector into the third risk prediction model and output the third sarcopenia probability value. The risk prediction value determination module is used to perform a weighted summation of the first sarcopenia probability value, the second sarcopenia probability value, and the third sarcopenia probability value, and then determine the maximum value in the weighted summation result as the sarcopenia risk prediction value.

9. The system according to any one of claims 1 to 8, characterized in that, The risk assessment module includes: The confidence calculation module is used to calculate the confidence level based on the sarcopenia risk prediction value and the reliability index. The risk assessment output module is used to output the sarcopenia risk assessment result based on the confidence level and the sarcopenia risk prediction value.

10. A non-contact sarcopenia risk assessment method based on multimodal thermal imaging, characterized in that, Includes the following steps: Thermal and RGB videos of the subjects were acquired during multiple motion phases. Key point detection and region of interest localization are performed on the thermal imaging video and the RGB video to obtain the first quadriceps femoris region of interest data. Then, the first quadriceps femoris region of interest data is subjected to quality inspection and motion compensation to obtain the second quadriceps femoris region of interest data. Temporal features are extracted from the region of interest data of the second quadriceps femoris muscle to obtain a temporal feature vector, and the reliability of the region of interest data of the second quadriceps femoris muscle is calculated to obtain a reliability index. The time-series feature vector is input into multiple pre-trained sarcopenia risk prediction models to obtain sarcopenia risk prediction values. Based on the sarcopenia risk prediction values ​​and the reliability index, the sarcopenia risk assessment result is obtained. The time-series feature vector includes the baseline temperature mean, maximum temperature rise, recovery half-life, metabolic recovery index, and thermal metabolic efficiency.

Images