Smartphone-Derived Joint Angular Velocities in Sit-to-Stand Motion: A Novel Spatiotemporal Marker for Symptomatic Knee Osteoarthritis

Smartphone-Derived Joint Angular Velocities in Sit-to-Stand Motion: A Novel Spatiotemporal Marker for Symptomatic Knee Osteoarthritis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Smartphone-Derived Joint Angular Velocities in Sit-to-Stand Motion: A Novel Spatiotemporal Marker for Symptomatic Knee Osteoarthritis CHUNYI WEN This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6225566/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Background and Objective: Knee osteoarthritis (OA) is a debilitating condition that compromises mobility and exacerbates knee pain, necessitating accurate and accessible diagnostic tools. Traditional motion capture technology, while effective, is often cost-prohibitive and limited to laboratory settings. In response, we developed a novel, smartphone-based approach utilizing spatiotemporal analysis of joint angular velocities and angles in sit-to-stand (STS) motion to detect symptomatic knee OA. Our deep learning model, STS-Dynamics Net, analyzed 864 sit-to-stand motion videos from 120 participants, providing a nuanced assessment of joint dynamics and temporal interactions in trunk, knee, and ankle angles and velocities. Notably, our findings demonstrate that joint angular velocities are a robust spatiotemporal biomarker for knee OA detection, outperforming the WOMAC questionnaire and maximum trunk angle in diagnostic accuracy and rivalling the performance of gold-standard 3D marker-based systems. Furthermore, our analysis revealed a significant correlation between angular velocities and muscle volumes and fat-to-muscle ratios in the quadriceps and hamstrings, underscoring the role of muscle weakness in knee OA pathogenesis. This innovative approach has the potential to revolutionize knee OA detection, enabling reliable, cost-effective, and self-administered assessments in community settings and bridging the gap in accessible healthcare monitoring. Health sciences/Rheumatology/Rheumatic diseases/Osteoarthritis Health sciences/Biomarkers/Diagnostic markers Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Knee osteoarthritis (OA) is a leading cause of disability in older adults, causing severe knee pain and impairing daily activities. Early-stage OA patients often exhibit functional impairments during everyday activities 1 . The sit-to-stand (STS) movement has been as one of the core set of performance test for OA patients 2 . The 30-second STS test, which involves counting the number of transitions, is a widely used method for assessing physical function. However, research has shown that kinematic analyses offer a more sensitive evaluation, revealing more detailed insights into movement dynamics beyond mere transition counts 3 – 5 . Marker-based motion analysis studies have demonstrated that angular velocities and range-of-motion of the joints are discriminating factors between individuals with knee OA and healthy controls 6 , 7 , underscoring their potential as critical markers for disease assessment. Despite the comprehensive measurements from marker-based 3D motion analysis, its high cost and specialized equipment limit its practicality 8 . Existing advancements in wearable sensors 9 , 10 and depth-camera systems unsupervised motion assessments more feasible. However, these technologies still require specialized hardware, limiting their widespread adoption. In this context, smartphones—ubiquitous and user-friendly—emerge as a promising alternative for accessible motion analysis. Recent studies have shown the potential of using static measures like maximum trunk angle derived from smartphone videos during STS movements to differentiate affected individuals from healthy controls 11 . However, these methods focus solely on predefined parameters of a single joint, failing to capture the temporal dynamics involving multiple joint coordination during the STS motion, resulting in a loss of information and an incomplete understanding of pathological movement. In response to these challenges, we aimed to develop an accessible and accurate diagnostic tool for knee OA by leveraging time-series joint angular velocities captured through ubiquitous smartphone technology. We performed a spatiotemporal analysis on angular velocities and joint angles of the adjacent joints from smartphone-captured sit-to-stand (STS) movements. This integration of angular velocities as spatiotemporal markers enriches our understanding of joint dynamics and patterns indicative of symptomatic knee OA. Building on this, we created STS-Dynamics Net, a deep learning model designed to detect autoregressive patterns and temporal interactions among the trunk, knee, and ankle joints during STS motions. This model not only improves the sensitivity of OA detection but also ensures scalability and accessibility. Further enhancing interpretability, we applied Gradient SHAP (Shapley Additive exPlanations) analysis 12 to reveal how individual joint metrics influence predictions. Additionally, our exploration of the link between joint dynamics and thigh muscle morphometry offers deep insights into biomechanical changes in knee OA. This holistic analysis enhances our understanding of knee OA and supports the development of an interpretable, precise, and accessible screening tool, poised to revolutionize screening and management of the condition. Results This study recruited 120 participants from Hong Kong, detailed in Table 1 , with an average age of 66.4 years (range: 45–81 years) and comprising 86 (71.7%) females. The average BMI was 23.0 kg/m² (SD = 3.31). Of these, 67 participants met the American College of Rheumatology (ACR) criteria for symptomatic knee OA in their right knees, while 53 were non-OA. The participants were split into training (n = 96) and testing (n = 24) groups for model development, with balanced demographic distribution. Table 1 Distributions of the demographics of the included subjects in training and testing sets. Parameter Whole Dataset Training Testing p -value No. of Subject 120 (100%) 96 (80%) 24 (20%) - No. of STS Videos 864 (100%) 681 (78.8%) 183 (21.2%) - KOA Diagnosis (ACR Criteria) Negative 53 42 11 0.6455 Positive 67 54 13 Age Range 45–81 45–81 49–76 0.2411 Mean 66.4 66.7 65.1 Std. Dev. 5.98 6.01 5.57 BMI Range 16.76–31.54 kg/m 2 16.8–31.5 kg/m 2 16.6–26.8 kg/m 2 0.024 Mean 23.0 kg/m 2 23.3 kg/m 2 21.6 kg/m 2 Std. Dev. 3.31 kg/m 2 3.28 kg/m 2 3.16 kg/m 2 Sex Male 34 (28.3%) 29 (30.2%) 5 (20.8%) 0.3619 Female 86 (71.7%) 67 (69.8%) 19 (79.2%) WOMAC Pain Score Range 0–18 0–18 0–15 0.9892 Median 2 2 1.5 IQR 4 4 4.25 WOMAC Stiffness Score Range 0–6 0–6 0–5 0.3810 Median 0 0 0 IQR 2 2 1 WOMAC Function Score Range 0–58 0–58 0–30 0.3006 Median 4 5 2.5 IQR 12.5 12.25 8.5 A total of 864 sit-to-stand (STS) motion videos were captured in the sagittal view. Using a pose estimation algorithm, joint centers for the right shoulder, hip, knee, ankle, and toe were identified in video frames, creating multivariate time series data of joint angles and velocities for the trunk, right knee, and ankle. This data fed into the development of STS-Dynamics Net, a deep learning classifier designed to distinguish between symptomatic knee OA and non-OA based on joint movement patterns (Fig. 2 ). Superiority of the STS-Dynamics Net The STS-Dynamics Net was specifically designed to model temporal dynamics, featuring a convolution module and a long-short-term memory (LSTM) recurrent module to effectively capture both local and global temporal features. Additionally, an attention layer was integrated to allow the model to focus on salient information crucial for discriminating between the classes. Our proposed model was benchmarked against two strong baseline architectures in the time series classification task, FCN 13 and LSTM-FCN 14 . Using joint angles and angular velocities as inputs, the STS-Dynamics Net achieved an AUC of the ROC curve (AUC-ROC) of 0.8484 ± 0.0324 and an AUC of the precision-recall curve (AUC-PR) of 0.8513 ± 0.0297, surpassing the FCN (0.7793 ± 0.0558 AUC-ROC; 0.7942 ± 0.0482 AUC-PR) and LSTM-FCN (0.7964 ± 0.0828 AUC-ROC; 0.8211 ± 0.0806 AUC-PR) in both metrics (refer to Supplementary Table 1 for details). Smartphone-derived joint angular velocity contributes to symptomatic knee OA identification Our marker-less sit-to-stand (STS) motion analysis model, using joint angles and angular velocities from smartphone videos, demonstrated substantial performance for identifying symptomatic knee osteoarthritis (OA) (Fig. 3 ). The model achieved an area under the ROC curve (AUC) of 0.8484 ± 0.0324, surpassing both the single-valued maximum trunk angle measurement (AUC = 0.6316 ± 0.0891) and the model using only joint angles (AUC = 0.8106 ± 0.0246), highlighting the diagnostic benefit of including angular velocities. However, adding angular accelerations did not significantly enhance performance (AUC = 0.8403 ± 0.0425) (Supplementary Table 2). Our method also rivals the performance of 3-dimensional marker-based motion capture (MoCap) systems (AUC = 0.8444 ± 0.0960) and proved to be cost-effective and accessible. Furthermore, it outperformed the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) questionnaire, a standard for assessing knee OA symptoms 15 , with a higher average AUC (0.7850 ± 0.1021) and greater robustness. Association between STS-D Index with WOMAC scores Besides comparing our model's performance with WOMAC scores for detecting symptomatic knee OA, we further examined how STS-D Index, which represents the model’s predicted probability of symptomatic knee OA (ranging from 0 to 1, with 1 representing the highest risk) relates to specific WOMAC sub-scores (Supplementary Fig. 2). Our findings reveal significant univariate correlations between the proposed index and the pain (rho = 0.53, p < 0.001), stiffness (rho = 0.45, p < 0.001), and function (rho = 0.62, p < 0.001) sub-scores, as well as the overall WOMAC score (rho = 0.62, p < 0.001). To manage potential spurious correlations arising from high co-linearity among WOMAC sub-scores, we conducted a multivariate regression analysis between these sub-scores and the STS-D Index (see Supplementary Table 3). The analysis revealed that only the function score from WOMAC showed a significant correlation (coefficient = 0.767, p < 0.001) with our index. The associations with pain (coefficient = -0.038, p = 0.765) and stiffness (coefficient = 0.101, p = 0.230) were found to be non-significant. Interpretability of the model To demystify the 'black box' of our deep neural network model, we employed gradient SHAP values to interpret its decision-making process and validate its clinical relevance. Our findings reveal that the model assigns importance to specific temporal phases across joint angles and angular velocities (Fig. 4 ). Notably, during the momentum shift and extension phase (phase ii), individuals with symptomatic OA exhibit smaller trunk flexion angles (p < 0.01) and a sharper peak in trunk extension angles compared to those without OA. The model also highlights significant differences in knee and ankle extensions, with the non-OA group achieving greater maximum knee extension (p < 0.01). Furthermore, during the controlled descent (phase iii), the OA group shows higher maximum trunk flexion angular velocity (p < 0.05), while the non-OA group exhibits higher knee angular velocities across both phases (p < 0.001). Video-based STS motion parameters associate with muscle mass and composition In our latest session, we continued our investigation into the motion parameters linked to recognizing symptomatic knee osteoarthritis (OA) by examining their associations with muscle morphometrics across various sit-to-stand (STS) phases. Using MRI, we manually segmented and analyzed the muscle volume and intramuscular fat-to-muscle volume ratio of the quadriceps and hamstrings, adjusting these volumes according to each subject's BMI. We focused especially on the rectus femoris (RF) and vastus medialis (VM) of the quadriceps, and the bicep femoris (BF) of the hamstrings. During phase ii (Momentum Shift & Full Extension), there was a significant negative correlation between the BF's normalized muscle volume and the trunk's maximum extension velocity (r = -0.442, p = 0.045), as shown in Fig. 5 a. Meanwhile, similar correlation was observed in the total hamstring normalized volume (r = -0.525, p = 0.018) as well (Supplementary Fig. 3). In phase iii (Controlled Descent), significant negative correlations were observed between both the quadriceps and hamstrings' normalized volumes and the trunk's maximum flexion velocity, notably for the RF (r = -0.482, p = 0.027) and BF (r = -0.632, p = 0.002). These results underscore the biomechanical influence of muscle mass on movement dynamics, indicating that greater muscle mass typically results in slower angular velocities during STS motions. Additionally, we explored the relationships between STS motion and intramuscular fat ratio (Fig. 5 b). During phase i (Flexion Momentum), a significant positive correlation was found between the VM's intramuscular fat ratio and the trunk's maximum flexion angular velocity (r = 0.615, p = 0.003). In phase ii, the BF's intramuscular fat ratio negatively correlated with both the knee's maximum extension angular velocity (r = -0.427, p = 0.054) and the trunk's minimum flexion angle during phase iii (r = -0.50, p = 0.021). Knee OA patients exhibit differentiated functional connectivity among truck, knee and ankle joints. Using the PCMCI + causality inference algorithm, we investigated the temporal functional connectivity of angular velocities between the trunk, knee, and ankle joints during specific phases of the sit-to-stand (STS) movement (Fig. 6 ): momentum shifts and full trunk extension (Phase ii), and controlled descent (Phase iii), involving simultaneous movement and coordination among the three joints. We noted distinct patterns of functional connectivity associated with ankle angular velocities among the OA and non-OA groups. Specifically, during Phase ii, symptomatic knee OA cases showed no significant connectivity between the joints, whereas healthy subjects displayed temporal connectivity extending from the knee to the ankle and then to the trunk. Conversely, in Phase iii, OA cases showed a directed connection from the ankle to the trunk, absent in healthy subjects who maintained a connection from the ankle to the knee instead. Additionally, the functional connection from the knee to the trunk is negatively influenced in healthy subjects but is positive in those with OA. These differences in connectivity patterns between angular velocities and joint angles, especially in OA cases, highlight the potential influence of knee dysfunction on joint coordination during dynamic activities. Discussion To our best knowledge, this study introduces the first smartphone video-derived joint angular velocities in sit-to-stand motions as a novel temporal marker for symptomatic knee OA. By capturing joint angular velocities through accessible smartphone video recordings, we provide a robust and innovative biomarker that enhances the detection of knee OA. To effectively analyze the complex dynamics inherent in this time-series data, we developed STS-Dynamics Net, a specialized deep-learning model designed to interpret the detailed temporal patterns of joint angles and angular velocities. We further defined the model’s predicted knee OA probability as STS-D Index, which achieved a disease detection performance with an area under the curve (AUC) of 0.8484 ± 0.0324. Furthermore, our single-camera, marker-less motion analysis approach demonstrated performance comparable to multi-camera, marker-based motion capture systems. Unlike existing methodologies that depend on wearable sensors such as accelerometers 16 , force plates 17 , or inertial measurement units 18 , 19 , our approach solely requires a readily available smartphone camera, offering significant advantages in scalability and ease of implementation for screening purpose. Our research emphasizes the crucial role of spatiotemporal analysis in assessing angular velocities of multiple connected joints for recognizing symptomatic knee osteoarthritis (OA) during sit-to-stand (STS) movements. Initially, our findings demonstrated that temporal velocity measurements not only surpassed the effectiveness of single-valued maximum trunk angle measurements but also showed enhanced value when integrated with temporal joint angle data. Beyond that, we conducted a functional connectivity analysis of the trunk, knee, and ankle joints, which uncovered significant deviations in the coordination patterns of angular velocities among subjects with knee OA. This indicates that individuals with knee OA may adopt modified synchronization strategies in their dynamic movements, likely as a compensatory mechanism to manage their condition 20 , 21 . These insights underscore the importance of employing a comprehensive dynamic analysis through our neural network model, which is adept at capturing not only the magnitude and sequential changes in joint kinematics but also the dynamic spatial interactions among the multiple joints of the lower limb 22 . Moreover, it merits further investigation to determine whether the connectivity patterns of angular velocities can be restored following rehabilitation, potentially serving as a marker for evaluating treatment outcomes for knee OA. We noted that increased trunk angular velocity is indicative of diminished muscle mass and a higher intramuscular fat ratio. Specifically, a pronounced negative correlation between trunk angular velocities and the muscle volumes of the quadriceps and hamstrings during various phases of the sit-to-stand (STS) movement was observed. This finding suggests a compensatory mechanism in which individuals with reduced muscle capacity—commonly seen in knee OA patients 23 , 24 — rely more heavily on trunk momentum to facilitate the STS movement. This adaptation allows patients to rise from the chair with reduced moments exerted at the knee joint 20 , 25 , 26 . Furthermore, an elevated intramuscular fat ratio in the quadriceps and hamstrings correlates with these trunk dynamics during flexion momentum and controlled descent phases, where fatty infiltration leads to muscle quality degradation and has been identified as a significant risk factor for the onset and progression of knee OA 27 – 29 . These observations highlight the potential of STS trunk dynamics as a valuable indicator of both the quantity and quality of thigh muscles in knee OA populations. To further evaluate the clinical relevance of our proposed STS-D Index, we compared it against the clinically prevalent WOMAC scores, and the discovered that the STS-D Index was closely associated with the WOMAC function and stiffness sub-scores, but not with the pain score. This finding aligns with simulation studies indicating that modifications in trunk motion are more closely tied to muscle weakness than to pain levels per se. Individuals with compromised muscle strength may exhibit altered or less efficient movement patterns that influence functional ability and perceived stiffness, whereas pain appears to have a less direct role in modulating trunk movement dynamics 30 . The study has several limitations. First, the small sample size and skewed sex distribution may limit the generalizability of our findings. Additionally, excluding individuals with low-back or hip pain, though essential for our focus, restricts broader applicability. Future studies should incorporate larger, more diverse cohorts, including those with these conditions, to enhance the system's comprehensiveness. Moreover, this study focused solely on detecting symptomatic OA in the right knee using sagittal plane videos. Given that unilateral OA results in asymmetrical body weight support 7 , the absence of frontal plane data prevents distinguishing between unilateral and bilateral OA, potentially limiting applicability for individuals with different knee conditions. Future improvements should explore multiple views 31 or marker-less three-dimensional pose estimation methods 32 to provide a more comprehensive knee analysis. Despite these limitations, our single-camera approach remains a promising, cost-effective tool for large-scale knee OA screening. In summary, this study introduces smartphone video-derived joint angular velocities as novel spatiotemporal markers for detecting symptomatic knee OA during sit-to-stand (STS) motions. We introduced the STS-D Index derived from a specifically designed deep learning model, STS-Dynamics Net that effectively analyzed the complex temporal dynamics of joint angles and velocities while offering performant disease detection capability. Additionally, by employing advanced model interpretation algorithm, we offered a nuanced understanding of knee kinematics associated with OA risk. Furthermore, the analysis revealed significant associations between angular velocities and both muscle volumes and fat-to-muscle ratios in the quadriceps and hamstrings, underscoring the role of muscle weakness in the pathogenesis of knee OA. Clinically, our approach presents a transformative tool for knee OA screening and ongoing self-monitoring, eliminating the need for costly laboratory equipment. The use of readily available smartphone cameras enables accessible, remote assessments in home and community settings, potentially facilitating large-scale screening and timely intervention. This advancement supports the shift towards decentralized healthcare, promoting cost-effective and user-friendly methods for managing knee OA. By leveraging smartphone technology, our method has the potential to enhance patient engagement, improve diagnostic accessibility, and ultimately contribute to better health outcomes for individuals with or at risk of knee osteoarthritis. Methods 1. Subject Recruitment The study received approval from The Hong Kong Polytechnic University, and all participants provided written informed consent. We initially screened 129 individuals but selected 120 based on specific criteria: age over 45, no persistent neck or lower back pain, no history of major joint surgery, no rheumatoid arthritis, able to perform the sit-to-stand (STS) motion, and having high-quality motion videos. The final participant group had an average age of 66.4 years (range 45–81), was 71.7% female, and had an average BMI of 23.0 kg/m 2 (range 16.76–31.54) (Table 1 ). Of these, 67 were diagnosed with symptomatic knee OA in the right knee according to the American College of Rheumatology (ACR) criteria, while the remaining 53 were classified as non-OA. Participants completed the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) questionnaire, which assesses pain, stiffness, and function in 24 questions scored on a 0–4 Likert scale, with higher scores indicating more severe symptoms. Total scores were calculated by summing responses for all sections. 2. Sit-to-stand Motion Video Acquisition and Marker-less Body Landmark Detection Participants initiated the sit-to-stand (STS) motion from a chair with adjustable height but no arms or backrest, starting with hands in front of the chest, hips and knees at about 90 degrees, and feet flat on the floor. After one or two practice repetitions, they performed the STS task at a self-selected speed and foot position for each of the 7 recorded trials on the same day, videos with poor quality were dropped. Motion videos were captured at 30 Hz and 1080p resolution using a smartphone on a tripod 2.4 m from the participant’s right side to record sagittal plane motions. Five 2D body landmarks (shoulder, hip, knee, ankle, and toe) on the right side were automatically identified in each frame using the BlazePose algorithm 33 . These landmarks underwent post-processing that included coordinate correction, gap filling, and smoothing with a fifth-order zero-lag low-pass Butterworth filter at 6 Hz, as per Boswell et al. 11 , and Gaussian smoothing (Fig. 2 a). Analysis included calculating three joint angles: trunk inclination, knee angle, and ankle angle, defined by specific body landmarks (Fig. 2 b). The resulting multivariate time series data were fitted with 3rd-order splines (smoothing factor of 80), and the first derivative was computed to determine the angular velocities for each joint angle (Fig. 2 a). 3. Model Architecture We developed the STS-Dynamics Net, a neural network designed to extract spatiotemporal information from motion data. The network consists of three 1D convolutional blocks with progressively decreasing kernel sizes (7, 5, and 1) and strides (3, 2, and 1) to maintain the input length throughout the layers 13 . Each block includes a Rectified Linear Unit (ReLU) activation 34 and 1D batch normalization to stabilize training. To address long-range temporal dependencies, an LSTM module follows the convolutional layers 35 . An Attention Pooling block is added after the LSTM to focus on key temporal sequences by calculating attention weights 36 . The outputs from this block are concatenated with globally pooled features from the convolutional layers, enhancing the network's ability to interpret both local and global temporal features 37 . Detailed network architecture is illustrated in Supplementary Fig. 1. Performance benchmarks were conducted against two baseline models: the Fully Convolutional Network (FCN) 13 and the LSTM-FCN 14 . FCN excels in extracting local temporal features, while LSTM-FCN integrates both local and long-range temporal dynamics. 4. Model Training Subjects were randomly divided into training and testing groups at an 8:2 ratio, producing 681 and 183 videos, respectively, from 7 trials per subject. Each motion trial was used as a separate sample for model development. To standardize input length for batch processing, multivariate sequence data were padded to 220. Data augmentations, including Gaussian noise, horizontal flipping, probabilistic cropping, random masking, and random average smoothing, were implemented to tackle overfitting. The model employed Binary Cross-Entropy loss, optimized using Adam with a learning rate of 0.001 and weight decay of 0.0138 38 . A dropout rate of 0.2 was applied to hidden layers to prevent overfitting further. The networks were developed using PyTorch v1.10.1, PyTorch-Lightning v1.9.0, and Python 3.7.5. 5. Marker-based Motion Capture Validation To compare our video-based marker-less STS motion analysis with conventional 3D marker-based motion capture. Together with the smartphone-taped motion videos, we also recorded the STS motion using a 10-camera Vicon system (Oxford Metrics, UK) at 200 Hz, tracking 39 retroreflective markers to map 3D marker trajectories. These trajectories were processed to calculate joint centers for the right shoulder, hip, knee, and ankle using the Dynamic Plug-in Gait Full-body Model, as per the Nexus 2 user guide. Both the marker-based and marker-less data underwent similar processing, including filtering through a 6 Hz, fifth-order, zero-lag, low-pass Butterworth filter. Joint angles for the trunk inclination, right knee, and right ankle were calculated, and their time series were fitted with 3rd-order splines to derive angular velocities, creating a six-channel multivariate time series for neural network input. The STS-Dynamics Net was then applied to both data sets using identical hyper-parameters for a balanced comparison. 6. Definition of STS Phases To conduct a detailed analysis of the Sit-to-Stand (STS) motion, we segmented it into four main phases based on trunk dynamics. Phase i encompasses the initial trunk forward lean until maximum hip flexion is achieved. Phase ii is characterized by a shift in momentum and hip extension, beginning at maximum trunk flexion and ending when the trunk is fully extended, signaling a stabilized standing posture. Phase iii involves a controlled descent where the trunk begins to lean forward again, lowering the body's center-of-mass until maximum degree of trunk flexion is reached, accompanied by knee flexion and ankle dorsiflexion. The final phase, Phase iv, involves extending the trunk from a fully flexed position back to an upright position to stabilize the sitting posture upon re-contact with the chair. 7. Interpretable Motion Deviations Between OA and non-OA Groups We utilized the Gradient SHapley Additive exPlanations (SHAP) 12 within a deep learning model interpretation framework to analyze motion patterns that differentiate diseased from non-diseased individuals, specifically focusing on knee OA. Gradient SHAP values, which were rescaled between 0 and 1, highlighted the importance of specific temporal segments in recognizing knee OA, with a higher value indicating greater significance. This analysis was conducted using Captum v0.4.0, a Python library based on PyTorch. For visual representation, we used the soft Dynamic Time Warping (DTW) Barycenter Averaging method to align and average the time series data, color-coding the results according to Gradient SHAP values. Further insights were gained by comparing time series segments with high normalized Gradient SHAP values between the OA and healthy control groups, using unpaired t-tests to statistically evaluate the differences. 8. Muscle Volume and Fat-to-muscle Volume Ratio Quantification on MRI We randomly selected 21 subjects to receive thigh magnetic resonance imaging (MRI). We acquired T1-weighted turbo spin echo (TSE) sequences of the entire thigh using a 3T MRI, with an in-plane resolution of 0.98 x 0.98 mm², a slice thickness of 5 mm, a matrix size of 512 x 256, and 90 slices. Preprocessing involved the N4ITK method for image normalization and field inhomogeneity correction. Manual segmentation of the full 3D volume was performed using ITKSnap, targeting the quadriceps, hamstrings, Rectus Femoris (RF), Vastus Medialis (VM), and Biceps Femoris (BF), with muscle volume calculated as the sum of all pixels in each segmentation 27 . The volumes are divided by the subjects’ BMI as for data normalization. Intramuscular adipose tissue (Intra-MAT) volume was determined via Otsu intensity thresholding 39 , with the ratio of Intra-MAT to muscle volume calculated for each muscle. 9. Directed Functional Connectivity of Angular Velocities We employed the PCMCI + algorithm 40 to uncover the temporal causal patterns among knee, trunk, and ankle angular velocities during STS motions in symptomatic OA and non-OA groups respectively. The algorithm was chosen for its robustness in the detection of causal links from high-dimensional time series data while mitigating false positives 41 , 42 . We performed 50 bootstrap samplings across all trials in the OA and healthy group separately. Within each bootstrap set, the motion trials were concatenated together to form a long sequence and is inferred by PCMCI + to identify statistically significant (p < 0.05) causal edges at specified time lags. These edges denote directional influences between the joints’ angular velocity time series. Subsequently, a set of consistent causal edges were identified if they remained significant in over 50% of all the bootstrapped samples, reflecting reproducibility. This procedure yielded a robust network of sequential connections among the joints. We displayed these consistent connections through both static and dynamic network graphs to provide a detailed view of the relationships. Specifically, we analyzed the temporal causal patterns in Phases ii and iii of the motion, where rapid coordination is observed among the three adjacent joints. 10. Statistical Analysis Subjects were randomly assigned to training and testing groups, with 96 subjects (80%) in the training set and 24 subjects (20%) in the testing set. Participant demographics were analyzed for homogeneity using an unpaired t-test for age and BMI, and Pearson's chi-squared test for sex and knee OA diagnosis. The performance of STS-D Index was evaluated using the receiver operating characteristic curve (ROC), the area under the ROC curve (AUC), and the area under the precision-recall curve (AUC-PR). Stratified bootstrapping with 50 iterations was used to calculate the mean and standard deviation of the AUC score. Correlation matrices were created to examine the relationships between joint kinematic parameters from Gradient SHAP analysis and muscle characteristics such as muscle volumes and the fat-to-muscle volume ratio in the Rectus Femoris (RF), Vastus Medialis (VM), and Biceps Femoris (BF). Motion parameters were extracted and averaged across each subject's STS trials. Pearson’s correlation coefficients were used to assess these correlations, which are depicted in bubble plots with color intensity showing correlation strength and bubble size indicating the p-value. Spearman’s correlation analysis was used to evaluate the consistency between the average STS-D Index, as determined by the STS-Dynamic Net, and WOMAC sub-section scores. We calculated the average predicted index for each participant by averaging the model predictions across all STS trials for that subject. Linear regression lines, with error bars representing the 95% confidence interval, were plotted to demonstrate the association trends. To mitigate potential spurious correlations due to co-linearity among the sub-scores, we conducted multivariate linear regression between the WOMAC scores and the STS-D Index. The scores were normalized to a 0–1 range before model-fitting, and significant regression coefficients were determined using Wald’s test. All statistical analyses were performed using the statistical extension of SciPy v1.14.1 library of Python 3.7.5. Declarations Acknowledgements We would like to express our gratitude to Yan Chai Hospital Social Service Department for their kind assistance in the subject referral, and to all the participants in the study. This work was supported by RISA seed fund (P0043002; P0051049; P0050709), RIAM seed fund (P0050824), Mainland / GBA Research Funding Scheme (P0049195), and Innovation & Technology Fund for Better Living (FBL/B046/22/S). Finally, we would like to appreciate Dr. Zhiqiang Wang for his valuable comments on the manuscript. Author contributions The concept and design of the study were developed by Lok Chun CHAN, Jin YAN, Dr. Ping Keung CHAN, Dr. Billy So, and Dr. Chunyi WEN. Data acquisition was carried out by Lok Chun CHAN, Jin YAN, Tianshu JIANG, and Ho Hin Toby LI, while data analysis and interpretation were conducted by Lok Chun CHAN, and Jin YAN. The initial draft of the manuscript was authored by Lok Chun CHAN. All authors contributed to revising the manuscript and offered scientific insights. The final manuscript received approval from all authors. Lok Chun CHAN, representing all contributors, is responsible for the entirety of the work, from its initiation to the finalized manuscript. Competing interests The authors have no competing interests to declare. References Mahmoudian, A., Lohmander, L. S., Mobasheri, A., Englund, M. & Luyten, F. P. Early-stage symptomatic osteoarthritis of the knee—time for action. Nature Reviews Rheumatology , 1-12 (2021). Dobson, F. et al. OARSI recommended performance-based tests to assess physical function in people diagnosed with hip or knee osteoarthritis. Osteoarthritis and cartilage 21 , 1042-1052 (2013). Turcot, K., Armand, S., Fritschy, D., Hoffmeyer, P. & Suvà, D. Sit-to-stand alterations in advanced knee osteoarthritis. Gait & posture 36 , 68-72 (2012). Millor, N., Lecumberri, P., Gomez, M., Martinez-Ramirez, A. & Izquierdo, M. Kinematic parameters to evaluate functional performance of sit-to-stand and stand-to-sit transitions using motion sensor devices: a systematic review. IEEE transactions on neural systems and rehabilitation engineering 22 , 926-936 (2014). Jørgensen, S. L., Mechlenburg, I., Bohn, M. B. & Aagaard, P. Sit-to-stand power predicts functional performance and patient-reported outcomes in patients with advanced knee osteoarthritis. A cross-sectional study. Musculoskeletal Science and Practice 69 , 102899 (2024). Pan, J. et al. Biomechanics of the lower limb in patients with mild knee osteoarthritis during the sit-to-stand task. BMC Musculoskeletal Disorders 25 , 268 (2024). Petrella, M. et al. Kinetics, kinematics, and knee muscle activation during sit to stand transition in unilateral and bilateral knee osteoarthritis. Gait & Posture 86 , 38-44 (2021). Kidziński, Ł. et al. Deep neural networks enable quantitative movement analysis using single-camera videos. Nature communications 11 , 1-10 (2020). Hellec, J., Chorin, F., Castagnetti, A. & Colson, S. S. Sit-to-stand movement evaluated using an inertial measurement unit embedded in smart glasses—A validation study. Sensors 20 , 5019 (2020). Ejupi, A. et al. Wavelet-based sit-to-stand detection and assessment of fall risk in older people using a wearable pendant device. IEEE Transactions on Biomedical Engineering 64 , 1602-1607 (2016). Boswell, M. A. et al. Smartphone videos of the sit-to-stand test predict osteoarthritis and health outcomes in a nationwide study. npj Digital Medicine 6 , 32 (2023). Lundberg, S. M. & Lee, S.-I. A unified approach to interpreting model predictions. Advances in neural information processing systems 30 (2017). Wang, Z., Yan, W. & Oates, T. in 2017 International joint conference on neural networks (IJCNN). 1578-1585 (IEEE). Karim, F., Majumdar, S., Darabi, H. & Chen, S. LSTM fully convolutional networks for time series classification. IEEE access 6 , 1662-1669 (2017). White, D. K. & Master, H. Patient-reported measures of physical function in knee osteoarthritis. Rheumatic Disease Clinics 42 , 239-252 (2016). Adamowicz, L. et al. Assessment of sit-to-stand transfers during daily life using an accelerometer on the lower back. Sensors 20 , 6618 (2020). Yamako, G., Chosa, E., Totoribe, K., Fukao, Y. & Deng, G. Quantification of the sit-to-stand movement for monitoring age-related motor deterioration using the Nintendo Wii Balance Board. PLoS one 12 , e0188165 (2017). Martinez-Hernandez, U. & Dehghani-Sanij, A. A. Probabilistic identification of sit-to-stand and stand-to-sit with a wearable sensor. Pattern Recognition Letters 118 , 32-41 (2019). Wairagkar, M. et al. A novel approach for modelling and classifying sit-to-stand kinematics using inertial sensors. Plos one 17 , e0264126 (2022). Anan, M. et al. Do patients with knee osteoarthritis perform sit-to-stand motion efficiently? Gait & posture 41 , 488-492 (2015). van Der Kruk, E., Silverman, A. K., Reilly, P. & Bull, A. M. Compensation due to age-related decline in sit-to-stand and sit-to-walk. Journal of biomechanics 122 , 110411 (2021). Tsang, M., Cheng, D. & Liu, Y. Detecting statistical interactions from neural network weights. arXiv preprint arXiv:1705.04977 (2017). Heiden, T. L., Lloyd, D. G. & Ackland, T. R. Knee extension and flexion weakness in people with knee osteoarthritis: is antagonist cocontraction a factor? journal of orthopaedic & sports physical therapy 39 , 807-815 (2009). Wu, Y. et al. Weight-Bearing Physical Activity, Lower-Limb Muscle Mass, and Risk of Knee Osteoarthritis. JAMA Network Open 7 , e248968-e248968 (2024). Doorenbosch, C. A., Harlaar, J., Roebroeck, M. E. & Lankhorst, G. J. Two strategies of transferring from sit-to-stand; the activation of monoarticular and biarticular muscles. Journal of biomechanics 27 , 1299-1307 (1994). Gismelseed, S., Al-Yahmedi, A., Zaier, R., Ouakad, H. & Bahadur, I. Predicting Sit-to-Stand Body Adaptation Using a Simple Model. Axioms 12 , 559 (2023). Mohajer, B. et al. Role of thigh muscle changes in knee osteoarthritis outcomes: osteoarthritis initiative data. Radiology 305 , 169-178 (2022). Pedroso, M. G., de Almeida, A. C., Aily, J. B., de Noronha, M. & Mattiello, S. M. Fatty infiltration in the thigh muscles in knee osteoarthritis: a systematic review and meta-analysis. Rheumatology international 39 , 627-635 (2019). Kumar, D. et al. Quadriceps intramuscular fat fraction rather than muscle size is associated with knee osteoarthritis. Osteoarthritis and cartilage 22 , 226-234 (2014). Van Der Kruk, E. & Geijtenbeek, T. Is increased trunk flexion in standing up related to muscle weakness or pain avoidance in individuals with unilateral knee pain; a simulation study. Frontiers in Bioengineering and Biotechnology 12 , 1346365 (2024). Keilstrup Ingwersen, C., Bjorholm Dahl, A., Nørtoft Jensen, J. & Rieger Hannemose, M. Two Views Are Better than One: Monocular 3D Pose Estimation with Multiview Consistency. arXiv e-prints , arXiv: 2311.12421 (2023). Zhu, W. et al. in Proceedings of the IEEE/CVF International Conference on Computer Vision. 15085-15099. Bazarevsky, V. et al. Blazepose: On-device real-time body pose tracking. arXiv preprint arXiv:2006.10204 (2020). Nair, V. & Hinton, G. Rectified Linear Units Improve Restricted Boltzmann Machines Vinod Nair . Vol. 27 (2010). Graves, A. & Graves, A. Long short-term memory. Supervised sequence labelling with recurrent neural networks , 37-45 (2012). Vinayavekhin, P. et al. in 2018 24th International Conference on Pattern Recognition (ICPR). 2624-2629 (IEEE). Dai, Y., Gieseke, F., Oehmcke, S., Wu, Y. & Barnard, K. in Proceedings of the IEEE/CVF winter conference on applications of computer vision. 3560-3569. Kingma, D. P. & Ba, J. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014). Ogier, A. C., Hostin, M.-A., Bellemare, M.-E. & Bendahan, D. Overview of MR image segmentation strategies in neuromuscular disorders. Frontiers in Neurology 12 , 625308 (2021). Runge, J. in Conference on Uncertainty in Artificial Intelligence. 1388-1397 (Pmlr). Runge, J. et al. (2019). Runge, J., Gerhardus, A., Varando, G., Eyring, V. & Camps-Valls, G. Causal inference for time series. Nature Reviews Earth & Environment 4 , 487-505 (2023). Additional Declarations There is NO Competing Interest. Supplementary Files SupplementoryInformation.docx Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6225566","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":439877882,"identity":"8bbd4bb9-c3a8-49c1-9a80-3496ccd75249","order_by":0,"name":"CHUNYI WEN","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7klEQVRIiWNgGAWjYDACCRBhACKYDzAwFDBAuIwNCcRoYUuAMojSAgY8BsRp4Z/dY/aYp4AhsV+659uDDwaHExvYm7dJMO5Iw23JnTPmxkDzE2fOObvdcAZIC8+xMgnGMzk4tRhI5JhJA7XkbriRuw3IAGoBikgwtlUQ1rL/Rs4ziBb5N0Rq2SCRwwa1hQekBbfDJG6klUnOMZCon3EjzRzol3TjNp60YovENtze55+RvE3izR8bYyDj2YMPFday/eyHN9742JaMUwsIMPFAYocNiJvBJEMCXg0MDIw/IDRIcR0BtaNgFIyCUTASAQBBD0125VkoFgAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0003-1949-7822","institution":"The Hong Kong Polytechnic University","correspondingAuthor":true,"prefix":"","firstName":"CHUNYI","middleName":"","lastName":"WEN","suffix":""}],"badges":[],"createdAt":"2025-03-14 11:06:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6225566/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6225566/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":80242146,"identity":"58152d5c-f762-4c58-a447-4001f160cdba","added_by":"auto","created_at":"2025-04-09 15:04:54","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":270877,"visible":true,"origin":"","legend":"\u003cp\u003eA prospective cohort of 129participants was enrolled in this study. Following rigorous exclusion criteria, including age restrictions, prior knee and hip arthroplasty, cervical and lumbar spine pain, rheumatoid arthritis diagnosis, significant motion limitations, and suboptimal video quality, the entire cohort of 120 subjects remained eligible for analysis. Notably, 67 participants met the American College of Rheumatology (ACR) diagnostic criteria for symptomatic knee OA, whereas 53individuals were classified as non-OA controls.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6225566/v1/5fa5b8aba057077367fb2418.png"},{"id":80242148,"identity":"8e2e7f2e-f2db-4cfc-8b00-67338ef3516f","added_by":"auto","created_at":"2025-04-09 15:04:54","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":345251,"visible":true,"origin":"","legend":"\u003cp\u003eKinematic analysis workflow for sit-to-stand (STS) motion. (a) Schematic representation of the proposed methodology, outlining the sequential steps involved in data processing and analysis. (b) Anatomical illustration of the STS movement, highlighting the key joint angles measured during the transition from sitting to standing.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6225566/v1/41c1b054e406c58a6f7572f1.png"},{"id":80243349,"identity":"c9f66201-14b4-4c74-a69e-18e10c8bc299","added_by":"auto","created_at":"2025-04-09 15:20:54","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":119758,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver Operating Characteristic (ROC) curves comparing the performance STS-D Index derived from our marker-less sit-to-stand motion analysis model, which incorporates the full temporal dynamics of joint angles (trunk, knee, and ankle) and angular velocities (AUC = 0.8484 ± 0.0324), to four benchmark models: (a) 3-dimensional marker-based MoCap system (AUC = 0.8444 ± 0.0960), (b) single-valued maximum trunk angle measurement (AUC = 0.6316 ± 0.0891), and (c) Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) scores (AUC = 0.7850 ± 0.1021).\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6225566/v1/652125347bf582161b1726bb.png"},{"id":80242147,"identity":"6f738ac4-b136-483e-b129-c76913694536","added_by":"auto","created_at":"2025-04-09 15:04:54","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":378233,"visible":true,"origin":"","legend":"\u003cp\u003eTime-series analysis of sit-to-stand motion patterns in knee osteoarthritis (OA) and healthy individuals. Average joint angles and angular velocities are presented as line plots, computed using the soft Dynamic Time Warping (DTW) Barycenter Averaging method. The plots are color-coded according to Gradient SHAP values, which represent the relative importance of each time-series segment towards OA detection. Higher values indicate greater contribution to OA classification. The sit-to-stand motion is segmented into four distinct phases: (i) trunk flexion, (ii) momentum transfer and hip extension (iii) controlled descent, (iv) seat contact and stabilization. We compared the minimum and maximum joint angles and angular velocities between OA and healthy groups in time-series segments with high normalized Gradient SHAP values. Statistical significance was assessed using unpaired t-tests, with p-values indicated as *p \u0026lt; 0.05, **p \u0026lt; 0.01, ***p \u0026lt; 0.001, and ns (non-significant).\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6225566/v1/521e5b72f92d0fc93644315e.png"},{"id":80242153,"identity":"f11b8942-a78f-40a8-a74e-bc1869fb9b8f","added_by":"auto","created_at":"2025-04-09 15:04:54","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":356970,"visible":true,"origin":"","legend":"\u003cp\u003eAssociation analysis of joint kinematic parameters and muscle morphometry in the thigh. Correlation matrices were constructed to investigate the relationships between joint kinematic parameters and (a) muscle volumes, as well as (b) intra-muscular fat-to-muscle volume ratios in the Rectus Femoris (RF), Vastus Medialis (VM), and Bicep Femoris (BF) muscles, derived from T1-weighted Double-Echo Steady-State (DESS) magnetic resonance imaging (MRI) of the thigh. Pearson's correlation coefficients were used to quantify the strength and direction of these relationships. The results are visualized as heatmaps, where the color intensity represents the magnitude of the correlation, with red and blue indicating positive and negative correlations, while size of the bubbles represents the degree of statistical significance, respectively. Statistical significance is denoted by *p \u0026lt; 0.05 and **p \u0026lt; 0.01.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6225566/v1/c9e201b916056d8552bce775.png"},{"id":80242160,"identity":"c70d6731-8912-447c-8baf-b8a40b72f6b6","added_by":"auto","created_at":"2025-04-09 15:04:54","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":192522,"visible":true,"origin":"","legend":"\u003cp\u003eDirected functional connectivity of the STS motion joint angular velocities in (a) phases ii and (b) iii. The network diagrams illustrate the sequential connectivity evaluated by PCMCI+ algorithm, where nodes represent angular velocities. The directed edges indicate statistically significant (p \u0026lt; 0.05) and consistent relationships across 50 bootstrapped samples. The arrow denotes temporal dependencies and information flow between joint angles and angular velocities during the sit-to-stand motion. The edges are color-coded based on causality value, with the intensity of the color reflecting the magnitude of causal strength. Negative and positive connectivities are depicted in blue and red, respectively, grey edges represent non-temporal associations between angular velocities\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6225566/v1/bb70ea6dbf2ed09256c7e18f.png"},{"id":80244247,"identity":"102e2ed0-bd6f-4045-b061-6ed4929a00b6","added_by":"auto","created_at":"2025-04-09 15:28:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2366440,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6225566/v1/ee74d26b-13ef-466c-bca6-55769cf033df.pdf"},{"id":80242154,"identity":"6958ec23-1bfa-428e-a13e-50c03a821eb1","added_by":"auto","created_at":"2025-04-09 15:04:54","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":668670,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementoryInformation.docx","url":"https://assets-eu.researchsquare.com/files/rs-6225566/v1/0d09230cd1db8a55d656cb5f.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Smartphone-Derived Joint Angular Velocities in Sit-to-Stand Motion: A Novel Spatiotemporal Marker for Symptomatic Knee Osteoarthritis","fulltext":[{"header":"Introduction","content":"\u003cp\u003eKnee osteoarthritis (OA) is a leading cause of disability in older adults, causing severe knee pain and impairing daily activities. Early-stage OA patients often exhibit functional impairments during everyday activities\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe sit-to-stand (STS) movement has been as one of the core set of performance test for OA patients\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. The 30-second STS test, which involves counting the number of transitions, is a widely used method for assessing physical function. However, research has shown that kinematic analyses offer a more sensitive evaluation, revealing more detailed insights into movement dynamics beyond mere transition counts\u003csup\u003e\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eMarker-based motion analysis studies have demonstrated that angular velocities and range-of-motion of the joints are discriminating factors between individuals with knee OA and healthy controls\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, underscoring their potential as critical markers for disease assessment. Despite the comprehensive measurements from marker-based 3D motion analysis, its high cost and specialized equipment limit its practicality\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. Existing advancements in wearable sensors\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e and depth-camera systems unsupervised motion assessments more feasible. However, these technologies still require specialized hardware, limiting their widespread adoption. In this context, smartphones\u0026mdash;ubiquitous and user-friendly\u0026mdash;emerge as a promising alternative for accessible motion analysis.\u003c/p\u003e \u003cp\u003eRecent studies have shown the potential of using static measures like maximum trunk angle derived from smartphone videos during STS movements to differentiate affected individuals from healthy controls\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. However, these methods focus solely on predefined parameters of a single joint, failing to capture the temporal dynamics involving multiple joint coordination during the STS motion, resulting in a loss of information and an incomplete understanding of pathological movement.\u003c/p\u003e \u003cp\u003eIn response to these challenges, we aimed to develop an accessible and accurate diagnostic tool for knee OA by leveraging time-series joint angular velocities captured through ubiquitous smartphone technology. We performed a spatiotemporal analysis on angular velocities and joint angles of the adjacent joints from smartphone-captured sit-to-stand (STS) movements. This integration of angular velocities as spatiotemporal markers enriches our understanding of joint dynamics and patterns indicative of symptomatic knee OA. Building on this, we created STS-Dynamics Net, a deep learning model designed to detect autoregressive patterns and temporal interactions among the trunk, knee, and ankle joints during STS motions. This model not only improves the sensitivity of OA detection but also ensures scalability and accessibility. Further enhancing interpretability, we applied Gradient SHAP (Shapley Additive exPlanations) analysis\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e to reveal how individual joint metrics influence predictions. Additionally, our exploration of the link between joint dynamics and thigh muscle morphometry offers deep insights into biomechanical changes in knee OA. This holistic analysis enhances our understanding of knee OA and supports the development of an interpretable, precise, and accessible screening tool, poised to revolutionize screening and management of the condition.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThis study recruited 120 participants from Hong Kong, detailed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, with an average age of 66.4 years (range: 45\u0026ndash;81 years) and comprising 86 (71.7%) females. The average BMI was 23.0 kg/m\u0026sup2; (SD\u0026thinsp;=\u0026thinsp;3.31). Of these, 67 participants met the American College of Rheumatology (ACR) criteria for symptomatic knee OA in their right knees, while 53 were non-OA. The participants were split into training (n\u0026thinsp;=\u0026thinsp;96) and testing (n\u0026thinsp;=\u0026thinsp;24) groups for model development, with balanced demographic distribution.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDistributions of the demographics of the included subjects in training and testing sets.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameter\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWhole Dataset\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTraining\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTesting\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo. of Subject\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e120 (100%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e96 (80%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e24 (20%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo. of STS Videos\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e864 (100%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e681 (78.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e183 (21.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eKOA Diagnosis\u003c/p\u003e \u003cp\u003e(ACR Criteria)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNegative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.6455\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRange\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45\u0026ndash;81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e45\u0026ndash;81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e49\u0026ndash;76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.2411\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e66.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e66.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e65.1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStd. Dev.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5.57\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eBMI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRange\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16.76\u0026ndash;31.54 kg/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e16.8\u0026ndash;31.5 kg/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e16.6\u0026ndash;26.8 kg/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.024\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e23.0 kg/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e23.3 kg/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e21.6 kg/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStd. Dev.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.31 kg/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.28 kg/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3.16 kg/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSex\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e34 (28.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e29 (30.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5 (20.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.3619\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e86 (71.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e67 (69.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e19 (79.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eWOMAC Pain Score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRange\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u0026ndash;18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0\u0026ndash;18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u0026ndash;15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.9892\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedian\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIQR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eWOMAC Stiffness Score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRange\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u0026ndash;6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0\u0026ndash;6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u0026ndash;5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.3810\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedian\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIQR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eWOMAC Function Score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRange\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u0026ndash;58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0\u0026ndash;58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u0026ndash;30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.3006\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedian\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIQR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e8.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eA total of 864 sit-to-stand (STS) motion videos were captured in the sagittal view. Using a pose estimation algorithm, joint centers for the right shoulder, hip, knee, ankle, and toe were identified in video frames, creating multivariate time series data of joint angles and velocities for the trunk, right knee, and ankle. This data fed into the development of STS-Dynamics Net, a deep learning classifier designed to distinguish between symptomatic knee OA and non-OA based on joint movement patterns (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eSuperiority of the STS-Dynamics Net\u003c/h2\u003e \u003cp\u003eThe STS-Dynamics Net was specifically designed to model temporal dynamics, featuring a convolution module and a long-short-term memory (LSTM) recurrent module to effectively capture both local and global temporal features. Additionally, an attention layer was integrated to allow the model to focus on salient information crucial for discriminating between the classes. Our proposed model was benchmarked against two strong baseline architectures in the time series classification task, FCN\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e and LSTM-FCN\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. Using joint angles and angular velocities as inputs, the STS-Dynamics Net achieved an AUC of the ROC curve (AUC-ROC) of 0.8484\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0324 and an AUC of the precision-recall curve (AUC-PR) of 0.8513\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0297, surpassing the FCN (0.7793\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0558 AUC-ROC; 0.7942\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0482 AUC-PR) and LSTM-FCN (0.7964\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0828 AUC-ROC; 0.8211\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0806 AUC-PR) in both metrics (refer to Supplementary Table\u0026nbsp;1 for details).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eSmartphone-derived joint angular velocity contributes to symptomatic knee OA identification\u003c/h3\u003e\n\u003cp\u003eOur marker-less sit-to-stand (STS) motion analysis model, using joint angles and angular velocities from smartphone videos, demonstrated substantial performance for identifying symptomatic knee osteoarthritis (OA) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The model achieved an area under the ROC curve (AUC) of 0.8484\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0324, surpassing both the single-valued maximum trunk angle measurement (AUC\u0026thinsp;=\u0026thinsp;0.6316\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0891) and the model using only joint angles (AUC\u0026thinsp;=\u0026thinsp;0.8106\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0246), highlighting the diagnostic benefit of including angular velocities. However, adding angular accelerations did not significantly enhance performance (AUC\u0026thinsp;=\u0026thinsp;0.8403\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0425) (Supplementary Table\u0026nbsp;2).\u003c/p\u003e \u003cp\u003eOur method also rivals the performance of 3-dimensional marker-based motion capture (MoCap) systems (AUC\u0026thinsp;=\u0026thinsp;0.8444\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0960) and proved to be cost-effective and accessible. Furthermore, it outperformed the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) questionnaire, a standard for assessing knee OA symptoms \u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, with a higher average AUC (0.7850\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1021) and greater robustness.\u003c/p\u003e\n\u003ch3\u003eAssociation between STS-D Index with WOMAC scores\u003c/h3\u003e\n\u003cp\u003eBesides comparing our model's performance with WOMAC scores for detecting symptomatic knee OA, we further examined how STS-D Index, which represents the model\u0026rsquo;s predicted probability of symptomatic knee OA (ranging from 0 to 1, with 1 representing the highest risk) relates to specific WOMAC sub-scores (Supplementary Fig.\u0026nbsp;2). Our findings reveal significant univariate correlations between the proposed index and the pain (rho\u0026thinsp;=\u0026thinsp;0.53, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), stiffness (rho\u0026thinsp;=\u0026thinsp;0.45, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and function (rho\u0026thinsp;=\u0026thinsp;0.62, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) sub-scores, as well as the overall WOMAC score (rho\u0026thinsp;=\u0026thinsp;0.62, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). To manage potential spurious correlations arising from high co-linearity among WOMAC sub-scores, we conducted a multivariate regression analysis between these sub-scores and the STS-D Index (see Supplementary Table\u0026nbsp;3). The analysis revealed that only the function score from WOMAC showed a significant correlation (coefficient\u0026thinsp;=\u0026thinsp;0.767, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) with our index. The associations with pain (coefficient = -0.038, p\u0026thinsp;=\u0026thinsp;0.765) and stiffness (coefficient\u0026thinsp;=\u0026thinsp;0.101, p\u0026thinsp;=\u0026thinsp;0.230) were found to be non-significant.\u003c/p\u003e\n\u003ch3\u003eInterpretability of the model\u003c/h3\u003e\n\u003cp\u003eTo demystify the 'black box' of our deep neural network model, we employed gradient SHAP values to interpret its decision-making process and validate its clinical relevance. Our findings reveal that the model assigns importance to specific temporal phases across joint angles and angular velocities (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Notably, during the momentum shift and extension phase (phase ii), individuals with symptomatic OA exhibit smaller trunk flexion angles (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) and a sharper peak in trunk extension angles compared to those without OA. The model also highlights significant differences in knee and ankle extensions, with the non-OA group achieving greater maximum knee extension (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Furthermore, during the controlled descent (phase iii), the OA group shows higher maximum trunk flexion angular velocity (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), while the non-OA group exhibits higher knee angular velocities across both phases (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\n\u003ch3\u003eVideo-based STS motion parameters associate with muscle mass and composition\u003c/h3\u003e\n\u003cp\u003eIn our latest session, we continued our investigation into the motion parameters linked to recognizing symptomatic knee osteoarthritis (OA) by examining their associations with muscle morphometrics across various sit-to-stand (STS) phases. Using MRI, we manually segmented and analyzed the muscle volume and intramuscular fat-to-muscle volume ratio of the quadriceps and hamstrings, adjusting these volumes according to each subject's BMI. We focused especially on the rectus femoris (RF) and vastus medialis (VM) of the quadriceps, and the bicep femoris (BF) of the hamstrings.\u003c/p\u003e \u003cp\u003eDuring phase ii (Momentum Shift \u0026amp; Full Extension), there was a significant negative correlation between the BF's normalized muscle volume and the trunk's maximum extension velocity (r = -0.442, p\u0026thinsp;=\u0026thinsp;0.045), as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea. Meanwhile, similar correlation was observed in the total hamstring normalized volume (r = -0.525, p\u0026thinsp;=\u0026thinsp;0.018) as well (Supplementary Fig.\u0026nbsp;3). In phase iii (Controlled Descent), significant negative correlations were observed between both the quadriceps and hamstrings' normalized volumes and the trunk's maximum flexion velocity, notably for the RF (r = -0.482, p\u0026thinsp;=\u0026thinsp;0.027) and BF (r = -0.632, p\u0026thinsp;=\u0026thinsp;0.002). These results underscore the biomechanical influence of muscle mass on movement dynamics, indicating that greater muscle mass typically results in slower angular velocities during STS motions.\u003c/p\u003e \u003cp\u003eAdditionally, we explored the relationships between STS motion and intramuscular fat ratio (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb). During phase i (Flexion Momentum), a significant positive correlation was found between the VM's intramuscular fat ratio and the trunk's maximum flexion angular velocity (r\u0026thinsp;=\u0026thinsp;0.615, p\u0026thinsp;=\u0026thinsp;0.003). In phase ii, the BF's intramuscular fat ratio negatively correlated with both the knee's maximum extension angular velocity (r = -0.427, p\u0026thinsp;=\u0026thinsp;0.054) and the trunk's minimum flexion angle during phase iii (r = -0.50, p\u0026thinsp;=\u0026thinsp;0.021).\u003c/p\u003e \u003cp\u003e \u003cb\u003eKnee OA patients exhibit differentiated functional connectivity among truck, knee and ankle joints.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eUsing the PCMCI\u0026thinsp;+\u0026thinsp;causality inference algorithm, we investigated the temporal functional connectivity of angular velocities between the trunk, knee, and ankle joints during specific phases of the sit-to-stand (STS) movement (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e): momentum shifts and full trunk extension (Phase ii), and controlled descent (Phase iii), involving simultaneous movement and coordination among the three joints. We noted distinct patterns of functional connectivity associated with ankle angular velocities among the OA and non-OA groups. Specifically, during Phase ii, symptomatic knee OA cases showed no significant connectivity between the joints, whereas healthy subjects displayed temporal connectivity extending from the knee to the ankle and then to the trunk. Conversely, in Phase iii, OA cases showed a directed connection from the ankle to the trunk, absent in healthy subjects who maintained a connection from the ankle to the knee instead. Additionally, the functional connection from the knee to the trunk is negatively influenced in healthy subjects but is positive in those with OA. These differences in connectivity patterns between angular velocities and joint angles, especially in OA cases, highlight the potential influence of knee dysfunction on joint coordination during dynamic activities.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eTo our best knowledge, this study introduces the first smartphone video-derived joint angular velocities in sit-to-stand motions as a novel temporal marker for symptomatic knee OA. By capturing joint angular velocities through accessible smartphone video recordings, we provide a robust and innovative biomarker that enhances the detection of knee OA. To effectively analyze the complex dynamics inherent in this time-series data, we developed STS-Dynamics Net, a specialized deep-learning model designed to interpret the detailed temporal patterns of joint angles and angular velocities. We further defined the model’s predicted knee OA probability as STS-D Index, which achieved a disease detection performance with an area under the curve (AUC) of 0.8484 ± 0.0324. Furthermore, our single-camera, marker-less motion analysis approach demonstrated performance comparable to multi-camera, marker-based motion capture systems. Unlike existing methodologies that depend on wearable sensors such as accelerometers \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e, force plates \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e, or inertial measurement units \u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e, our approach solely requires a readily available smartphone camera, offering significant advantages in scalability and ease of implementation for screening purpose.\u003c/p\u003e \u003cp\u003eOur research emphasizes the crucial role of spatiotemporal analysis in assessing angular velocities of multiple connected joints for recognizing symptomatic knee osteoarthritis (OA) during sit-to-stand (STS) movements. Initially, our findings demonstrated that temporal velocity measurements not only surpassed the effectiveness of single-valued maximum trunk angle measurements but also showed enhanced value when integrated with temporal joint angle data. Beyond that, we conducted a functional connectivity analysis of the trunk, knee, and ankle joints, which uncovered significant deviations in the coordination patterns of angular velocities among subjects with knee OA. This indicates that individuals with knee OA may adopt modified synchronization strategies in their dynamic movements, likely as a compensatory mechanism to manage their condition \u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e,\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. These insights underscore the importance of employing a comprehensive dynamic analysis through our neural network model, which is adept at capturing not only the magnitude and sequential changes in joint kinematics but also the dynamic spatial interactions among the multiple joints of the lower limb \u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Moreover, it merits further investigation to determine whether the connectivity patterns of angular velocities can be restored following rehabilitation, potentially serving as a marker for evaluating treatment outcomes for knee OA.\u003c/p\u003e \u003cp\u003eWe noted that increased trunk angular velocity is indicative of diminished muscle mass and a higher intramuscular fat ratio. Specifically, a pronounced negative correlation between trunk angular velocities and the muscle volumes of the quadriceps and hamstrings during various phases of the sit-to-stand (STS) movement was observed. This finding suggests a compensatory mechanism in which individuals with reduced muscle capacity—commonly seen in knee OA patients\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e,\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e — rely more heavily on trunk momentum to facilitate the STS movement. This adaptation allows patients to rise from the chair with reduced moments exerted at the knee joint\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e,\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. Furthermore, an elevated intramuscular fat ratio in the quadriceps and hamstrings correlates with these trunk dynamics during flexion momentum and controlled descent phases, where fatty infiltration leads to muscle quality degradation and has been identified as a significant risk factor for the onset and progression of knee OA\u003csup\u003e\u003cspan additionalcitationids=\"CR28\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e–\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. These observations highlight the potential of STS trunk dynamics as a valuable indicator of both the quantity and quality of thigh muscles in knee OA populations.\u003c/p\u003e \u003cp\u003eTo further evaluate the clinical relevance of our proposed STS-D Index, we compared it against the clinically prevalent WOMAC scores, and the discovered that the STS-D Index was closely associated with the WOMAC function and stiffness sub-scores, but not with the pain score. This finding aligns with simulation studies indicating that modifications in trunk motion are more closely tied to muscle weakness than to pain levels per se. Individuals with compromised muscle strength may exhibit altered or less efficient movement patterns that influence functional ability and perceived stiffness, whereas pain appears to have a less direct role in modulating trunk movement dynamics\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe study has several limitations. First, the small sample size and skewed sex distribution may limit the generalizability of our findings. Additionally, excluding individuals with low-back or hip pain, though essential for our focus, restricts broader applicability. Future studies should incorporate larger, more diverse cohorts, including those with these conditions, to enhance the system's comprehensiveness. Moreover, this study focused solely on detecting symptomatic OA in the right knee using sagittal plane videos. Given that unilateral OA results in asymmetrical body weight support\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, the absence of frontal plane data prevents distinguishing between unilateral and bilateral OA, potentially limiting applicability for individuals with different knee conditions. Future improvements should explore multiple views\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e or marker-less three-dimensional pose estimation methods \u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e to provide a more comprehensive knee analysis. Despite these limitations, our single-camera approach remains a promising, cost-effective tool for large-scale knee OA screening.\u003c/p\u003e \u003cp\u003eIn summary, this study introduces smartphone video-derived joint angular velocities as novel spatiotemporal markers for detecting symptomatic knee OA during sit-to-stand (STS) motions. We introduced the STS-D Index derived from a specifically designed deep learning model, STS-Dynamics Net that effectively analyzed the complex temporal dynamics of joint angles and velocities while offering performant disease detection capability. Additionally, by employing advanced model interpretation algorithm, we offered a nuanced understanding of knee kinematics associated with OA risk. Furthermore, the analysis revealed significant associations between angular velocities and both muscle volumes and fat-to-muscle ratios in the quadriceps and hamstrings, underscoring the role of muscle weakness in the pathogenesis of knee OA.\u003c/p\u003e \u003cp\u003eClinically, our approach presents a transformative tool for knee OA screening and ongoing self-monitoring, eliminating the need for costly laboratory equipment. The use of readily available smartphone cameras enables accessible, remote assessments in home and community settings, potentially facilitating large-scale screening and timely intervention. This advancement supports the shift towards decentralized healthcare, promoting cost-effective and user-friendly methods for managing knee OA. By leveraging smartphone technology, our method has the potential to enhance patient engagement, improve diagnostic accessibility, and ultimately contribute to better health outcomes for individuals with or at risk of knee osteoarthritis.\u003c/p\u003e \n\n \n\n "},{"header":"Methods","content":"\u003ch3\u003e1. Subject Recruitment\u003c/h3\u003e\u003cp\u003e The study received approval from The Hong Kong Polytechnic University, and all participants provided written informed consent. We initially screened 129 individuals but selected 120 based on specific criteria: age over 45, no persistent neck or lower back pain, no history of major joint surgery, no rheumatoid arthritis, able to perform the sit-to-stand (STS) motion, and having high-quality motion videos. The final participant group had an average age of 66.4 years (range 45–81), was 71.7% female, and had an average BMI of 23.0 kg/m\u003csup\u003e2\u003c/sup\u003e (range 16.76–31.54) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eOf these, 67 were diagnosed with symptomatic knee OA in the right knee according to the American College of Rheumatology (ACR) criteria, while the remaining 53 were classified as non-OA. Participants completed the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) questionnaire, which assesses pain, stiffness, and function in 24 questions scored on a 0–4 Likert scale, with higher scores indicating more severe symptoms. Total scores were calculated by summing responses for all sections.\u003c/p\u003e\u003ch3\u003e2. Sit-to-stand Motion Video Acquisition and Marker-less Body Landmark Detection\u003c/h3\u003e\u003cp\u003eParticipants initiated the sit-to-stand (STS) motion from a chair with adjustable height but no arms or backrest, starting with hands in front of the chest, hips and knees at about 90 degrees, and feet flat on the floor. After one or two practice repetitions, they performed the STS task at a self-selected speed and foot position for each of the 7 recorded trials on the same day, videos with poor quality were dropped.\u003c/p\u003e\u003cp\u003eMotion videos were captured at 30 Hz and 1080p resolution using a smartphone on a tripod 2.4 m from the participant’s right side to record sagittal plane motions. Five 2D body landmarks (shoulder, hip, knee, ankle, and toe) on the right side were automatically identified in each frame using the BlazePose algorithm \u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. These landmarks underwent post-processing that included coordinate correction, gap filling, and smoothing with a fifth-order zero-lag low-pass Butterworth filter at 6 Hz, as per Boswell et al. \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e, and Gaussian smoothing (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea).\u003c/p\u003e\u003cp\u003eAnalysis included calculating three joint angles: trunk inclination, knee angle, and ankle angle, defined by specific body landmarks (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb). The resulting multivariate time series data were fitted with 3rd-order splines (smoothing factor of 80), and the first derivative was computed to determine the angular velocities for each joint angle (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea).\u003c/p\u003e\u003ch2\u003e3. Model Architecture\u003c/h2\u003e\u003cp\u003eWe developed the STS-Dynamics Net, a neural network designed to extract spatiotemporal information from motion data. The network consists of three 1D convolutional blocks with progressively decreasing kernel sizes (7, 5, and 1) and strides (3, 2, and 1) to maintain the input length throughout the layers\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Each block includes a Rectified Linear Unit (ReLU) activation\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e and 1D batch normalization to stabilize training.\u003c/p\u003e\u003cp\u003eTo address long-range temporal dependencies, an LSTM module follows the convolutional layers\u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e. An Attention Pooling block is added after the LSTM to focus on key temporal sequences by calculating attention weights\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e. The outputs from this block are concatenated with globally pooled features from the convolutional layers, enhancing the network's ability to interpret both local and global temporal features\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. Detailed network architecture is illustrated in Supplementary Fig.\u0026nbsp;1.\u003c/p\u003e\u003cp\u003ePerformance benchmarks were conducted against two baseline models: the Fully Convolutional Network (FCN)\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e and the LSTM-FCN\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. FCN excels in extracting local temporal features, while LSTM-FCN integrates both local and long-range temporal dynamics.\u003c/p\u003e\u003ch2\u003e4. Model Training\u003c/h2\u003e\u003cp\u003eSubjects were randomly divided into training and testing groups at an 8:2 ratio, producing 681 and 183 videos, respectively, from 7 trials per subject. Each motion trial was used as a separate sample for model development. To standardize input length for batch processing, multivariate sequence data were padded to 220. Data augmentations, including Gaussian noise, horizontal flipping, probabilistic cropping, random masking, and random average smoothing, were implemented to tackle overfitting. The model employed Binary Cross-Entropy loss, optimized using Adam with a learning rate of 0.001 and weight decay of 0.0138\u003csup\u003e38\u003c/sup\u003e. A dropout rate of 0.2 was applied to hidden layers to prevent overfitting further. The networks were developed using PyTorch v1.10.1, PyTorch-Lightning v1.9.0, and Python 3.7.5.\u003c/p\u003e\u003ch2\u003e5. Marker-based Motion Capture Validation\u003c/h2\u003e\u003cp\u003eTo compare our video-based marker-less STS motion analysis with conventional 3D marker-based motion capture. Together with the smartphone-taped motion videos, we also recorded the STS motion using a 10-camera Vicon system (Oxford Metrics, UK) at 200 Hz, tracking 39 retroreflective markers to map 3D marker trajectories. These trajectories were processed to calculate joint centers for the right shoulder, hip, knee, and ankle using the Dynamic Plug-in Gait Full-body Model, as per the Nexus 2 user guide.\u003c/p\u003e\u003cp\u003eBoth the marker-based and marker-less data underwent similar processing, including filtering through a 6 Hz, fifth-order, zero-lag, low-pass Butterworth filter. Joint angles for the trunk inclination, right knee, and right ankle were calculated, and their time series were fitted with 3rd-order splines to derive angular velocities, creating a six-channel multivariate time series for neural network input. The STS-Dynamics Net was then applied to both data sets using identical hyper-parameters for a balanced comparison.\u003c/p\u003e\u003ch2\u003e6. Definition of STS Phases\u003c/h2\u003e\u003cp\u003eTo conduct a detailed analysis of the Sit-to-Stand (STS) motion, we segmented it into four main phases based on trunk dynamics. Phase i encompasses the initial trunk forward lean until maximum hip flexion is achieved. Phase ii is characterized by a shift in momentum and hip extension, beginning at maximum trunk flexion and ending when the trunk is fully extended, signaling a stabilized standing posture. Phase iii involves a controlled descent where the trunk begins to lean forward again, lowering the body's center-of-mass until maximum degree of trunk flexion is reached, accompanied by knee flexion and ankle dorsiflexion. The final phase, Phase iv, involves extending the trunk from a fully flexed position back to an upright position to stabilize the sitting posture upon re-contact with the chair.\u003c/p\u003e\u003ch2\u003e7. Interpretable Motion Deviations Between OA and non-OA Groups\u003c/h2\u003e\u003cp\u003eWe utilized the Gradient SHapley Additive exPlanations (SHAP)\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e within a deep learning model interpretation framework to analyze motion patterns that differentiate diseased from non-diseased individuals, specifically focusing on knee OA. Gradient SHAP values, which were rescaled between 0 and 1, highlighted the importance of specific temporal segments in recognizing knee OA, with a higher value indicating greater significance. This analysis was conducted using Captum v0.4.0, a Python library based on PyTorch.\u003c/p\u003e\u003cp\u003eFor visual representation, we used the soft Dynamic Time Warping (DTW) Barycenter Averaging method to align and average the time series data, color-coding the results according to Gradient SHAP values. Further insights were gained by comparing time series segments with high normalized Gradient SHAP values between the OA and healthy control groups, using unpaired t-tests to statistically evaluate the differences.\u003c/p\u003e\u003ch2\u003e8. Muscle Volume and Fat-to-muscle Volume Ratio Quantification on MRI\u003c/h2\u003e\u003cp\u003eWe randomly selected 21 subjects to receive thigh magnetic resonance imaging (MRI). We acquired T1-weighted turbo spin echo (TSE) sequences of the entire thigh using a 3T MRI, with an in-plane resolution of 0.98 x 0.98 mm², a slice thickness of 5 mm, a matrix size of 512 x 256, and 90 slices. Preprocessing involved the N4ITK method for image normalization and field inhomogeneity correction. Manual segmentation of the full 3D volume was performed using ITKSnap, targeting the quadriceps, hamstrings, Rectus Femoris (RF), Vastus Medialis (VM), and Biceps Femoris (BF), with muscle volume calculated as the sum of all pixels in each segmentation\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. The volumes are divided by the subjects’ BMI as for data normalization. Intramuscular adipose tissue (Intra-MAT) volume was determined via Otsu intensity thresholding\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e, with the ratio of Intra-MAT to muscle volume calculated for each muscle.\u003c/p\u003e\u003ch2\u003e9. Directed Functional Connectivity of Angular Velocities\u003c/h2\u003e\u003cp\u003eWe employed the PCMCI + algorithm\u003csup\u003e40\u003c/sup\u003e to uncover the temporal causal patterns among knee, trunk, and ankle angular velocities during STS motions in symptomatic OA and non-OA groups respectively. The algorithm was chosen for its robustness in the detection of causal links from high-dimensional time series data while mitigating false positives \u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e,\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e. We performed 50 bootstrap samplings across all trials in the OA and healthy group separately. Within each bootstrap set, the motion trials were concatenated together to form a long sequence and is inferred by PCMCI + to identify statistically significant (p \u0026lt; 0.05) causal edges at specified time lags. These edges denote directional influences between the joints’ angular velocity time series. Subsequently, a set of consistent causal edges were identified if they remained significant in over 50% of all the bootstrapped samples, reflecting reproducibility. This procedure yielded a robust network of sequential connections among the joints. We displayed these consistent connections through both static and dynamic network graphs to provide a detailed view of the relationships. Specifically, we analyzed the temporal causal patterns in Phases ii and iii of the motion, where rapid coordination is observed among the three adjacent joints.\u003c/p\u003e\u003ch2\u003e10. Statistical Analysis\u003c/h2\u003e\u003cp\u003eSubjects were randomly assigned to training and testing groups, with 96 subjects (80%) in the training set and 24 subjects (20%) in the testing set. Participant demographics were analyzed for homogeneity using an unpaired t-test for age and BMI, and Pearson's chi-squared test for sex and knee OA diagnosis.\u003c/p\u003e\u003cp\u003eThe performance of STS-D Index was evaluated using the receiver operating characteristic curve (ROC), the area under the ROC curve (AUC), and the area under the precision-recall curve (AUC-PR). Stratified bootstrapping with 50 iterations was used to calculate the mean and standard deviation of the AUC score.\u003c/p\u003e\u003cp\u003eCorrelation matrices were created to examine the relationships between joint kinematic parameters from Gradient SHAP analysis and muscle characteristics such as muscle volumes and the fat-to-muscle volume ratio in the Rectus Femoris (RF), Vastus Medialis (VM), and Biceps Femoris (BF). Motion parameters were extracted and averaged across each subject's STS trials. Pearson’s correlation coefficients were used to assess these correlations, which are depicted in bubble plots with color intensity showing correlation strength and bubble size indicating the p-value.\u003c/p\u003e\u003cp\u003eSpearman’s correlation analysis was used to evaluate the consistency between the average STS-D Index, as determined by the STS-Dynamic Net, and WOMAC sub-section scores. We calculated the average predicted index for each participant by averaging the model predictions across all STS trials for that subject. Linear regression lines, with error bars representing the 95% confidence interval, were plotted to demonstrate the association trends. To mitigate potential spurious correlations due to co-linearity among the sub-scores, we conducted multivariate linear regression between the WOMAC scores and the STS-D Index. The scores were normalized to a 0–1 range before model-fitting, and significant regression coefficients were determined using Wald’s test.\u003c/p\u003e\u003cp\u003eAll statistical analyses were performed using the statistical extension of SciPy v1.14.1 library of Python 3.7.5.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to express our gratitude to Yan Chai Hospital Social Service Department for their kind assistance in the subject referral, and to all the participants in the study. This work was supported by RISA seed fund (P0043002; P0051049; P0050709), RIAM seed fund (P0050824), Mainland / GBA Research Funding Scheme (P0049195), and Innovation \u0026amp; Technology Fund for Better Living (FBL/B046/22/S). Finally, we would like to appreciate Dr. Zhiqiang Wang for his valuable comments on the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe concept and design of the study were developed by Lok Chun CHAN, Jin YAN, Dr. Ping Keung CHAN, Dr. Billy So, and Dr. Chunyi WEN. Data acquisition was carried out by Lok Chun CHAN, Jin YAN, Tianshu JIANG, and Ho Hin Toby LI, while data analysis and interpretation were conducted by Lok Chun CHAN, and Jin YAN. The initial draft of the manuscript was authored by Lok Chun CHAN. All authors contributed to revising the manuscript and offered scientific insights. The final manuscript received approval from all authors. Lok Chun CHAN, representing all contributors, is responsible for the entirety of the work, from its initiation to the finalized manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no competing interests to declare.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eMahmoudian, A., Lohmander, L. S., Mobasheri, A., Englund, M. \u0026amp; Luyten, F. P. Early-stage symptomatic osteoarthritis of the knee\u0026mdash;time for action. \u003cem\u003eNature Reviews Rheumatology\u003c/em\u003e, 1-12 (2021).\u003c/li\u003e\n \u003cli\u003eDobson, F.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e OARSI recommended performance-based tests to assess physical function in people diagnosed with hip or knee osteoarthritis. \u003cem\u003eOsteoarthritis and cartilage\u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e, 1042-1052 (2013).\u003c/li\u003e\n \u003cli\u003eTurcot, K., Armand, S., Fritschy, D., Hoffmeyer, P. \u0026amp; Suv\u0026agrave;, D. Sit-to-stand alterations in advanced knee osteoarthritis. \u003cem\u003eGait \u0026amp; posture\u003c/em\u003e\u003cstrong\u003e36\u003c/strong\u003e, 68-72 (2012).\u003c/li\u003e\n \u003cli\u003eMillor, N., Lecumberri, P., Gomez, M., Martinez-Ramirez, A. \u0026amp; Izquierdo, M. Kinematic parameters to evaluate functional performance of sit-to-stand and stand-to-sit transitions using motion sensor devices: a systematic review. \u003cem\u003eIEEE transactions on neural systems and rehabilitation engineering\u003c/em\u003e\u003cstrong\u003e22\u003c/strong\u003e, 926-936 (2014).\u003c/li\u003e\n \u003cli\u003eJ\u0026oslash;rgensen, S. L., Mechlenburg, I., Bohn, M. B. \u0026amp; Aagaard, P. Sit-to-stand power predicts functional performance and patient-reported outcomes in patients with advanced knee osteoarthritis. A cross-sectional study. \u003cem\u003eMusculoskeletal Science and Practice\u003c/em\u003e\u003cstrong\u003e69\u003c/strong\u003e, 102899 (2024).\u003c/li\u003e\n \u003cli\u003ePan, J.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Biomechanics of the lower limb in patients with mild knee osteoarthritis during the sit-to-stand task. \u003cem\u003eBMC Musculoskeletal Disorders\u003c/em\u003e\u003cstrong\u003e25\u003c/strong\u003e, 268 (2024).\u003c/li\u003e\n \u003cli\u003ePetrella, M.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Kinetics, kinematics, and knee muscle activation during sit to stand transition in unilateral and bilateral knee osteoarthritis. \u003cem\u003eGait \u0026amp; Posture\u003c/em\u003e\u003cstrong\u003e86\u003c/strong\u003e, 38-44 (2021).\u003c/li\u003e\n \u003cli\u003eKidziński, Ł.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Deep neural networks enable quantitative movement analysis using single-camera videos. \u003cem\u003eNature communications\u003c/em\u003e\u003cstrong\u003e11\u003c/strong\u003e, 1-10 (2020).\u003c/li\u003e\n \u003cli\u003eHellec, J., Chorin, F., Castagnetti, A. \u0026amp; Colson, S. S. Sit-to-stand movement evaluated using an inertial measurement unit embedded in smart glasses\u0026mdash;A validation study. \u003cem\u003eSensors\u003c/em\u003e\u003cstrong\u003e20\u003c/strong\u003e, 5019 (2020).\u003c/li\u003e\n \u003cli\u003eEjupi, A.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Wavelet-based sit-to-stand detection and assessment of fall risk in older people using a wearable pendant device. \u003cem\u003eIEEE Transactions on Biomedical Engineering\u003c/em\u003e\u003cstrong\u003e64\u003c/strong\u003e, 1602-1607 (2016).\u003c/li\u003e\n \u003cli\u003eBoswell, M. A.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Smartphone videos of the sit-to-stand test predict osteoarthritis and health outcomes in a nationwide study. \u003cem\u003enpj Digital Medicine\u003c/em\u003e\u003cstrong\u003e6\u003c/strong\u003e, 32 (2023).\u003c/li\u003e\n \u003cli\u003eLundberg, S. M. \u0026amp; Lee, S.-I. A unified approach to interpreting model predictions. \u003cem\u003eAdvances in neural information processing systems\u003c/em\u003e\u003cstrong\u003e30\u003c/strong\u003e (2017).\u003c/li\u003e\n \u003cli\u003eWang, Z., Yan, W. \u0026amp; Oates, T. in \u003cem\u003e2017 International joint conference on neural networks (IJCNN).\u003c/em\u003e 1578-1585 (IEEE).\u003c/li\u003e\n \u003cli\u003eKarim, F., Majumdar, S., Darabi, H. \u0026amp; Chen, S. LSTM fully convolutional networks for time series classification. \u003cem\u003eIEEE access\u003c/em\u003e\u003cstrong\u003e6\u003c/strong\u003e, 1662-1669 (2017).\u003c/li\u003e\n \u003cli\u003eWhite, D. K. \u0026amp; Master, H. Patient-reported measures of physical function in knee osteoarthritis. \u003cem\u003eRheumatic Disease Clinics\u003c/em\u003e\u003cstrong\u003e42\u003c/strong\u003e, 239-252 (2016).\u003c/li\u003e\n \u003cli\u003eAdamowicz, L.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Assessment of sit-to-stand transfers during daily life using an accelerometer on the lower back. \u003cem\u003eSensors\u003c/em\u003e\u003cstrong\u003e20\u003c/strong\u003e, 6618 (2020).\u003c/li\u003e\n \u003cli\u003eYamako, G., Chosa, E., Totoribe, K., Fukao, Y. \u0026amp; Deng, G. Quantification of the sit-to-stand movement for monitoring age-related motor deterioration using the Nintendo Wii Balance Board. \u003cem\u003ePLoS one\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, e0188165 (2017).\u003c/li\u003e\n \u003cli\u003eMartinez-Hernandez, U. \u0026amp; Dehghani-Sanij, A. A. Probabilistic identification of sit-to-stand and stand-to-sit with a wearable sensor. \u003cem\u003ePattern Recognition Letters\u003c/em\u003e\u003cstrong\u003e118\u003c/strong\u003e, 32-41 (2019).\u003c/li\u003e\n \u003cli\u003eWairagkar, M.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e A novel approach for modelling and classifying sit-to-stand kinematics using inertial sensors. \u003cem\u003ePlos one\u003c/em\u003e\u003cstrong\u003e17\u003c/strong\u003e, e0264126 (2022).\u003c/li\u003e\n \u003cli\u003eAnan, M.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Do patients with knee osteoarthritis perform sit-to-stand motion efficiently? \u003cem\u003eGait \u0026amp; posture\u003c/em\u003e\u003cstrong\u003e41\u003c/strong\u003e, 488-492 (2015).\u003c/li\u003e\n \u003cli\u003evan Der Kruk, E., Silverman, A. K., Reilly, P. \u0026amp; Bull, A. M. Compensation due to age-related decline in sit-to-stand and sit-to-walk. \u003cem\u003eJournal of biomechanics\u003c/em\u003e\u003cstrong\u003e122\u003c/strong\u003e, 110411 (2021).\u003c/li\u003e\n \u003cli\u003eTsang, M., Cheng, D. \u0026amp; Liu, Y. Detecting statistical interactions from neural network weights. \u003cem\u003earXiv preprint arXiv:1705.04977\u003c/em\u003e (2017).\u003c/li\u003e\n \u003cli\u003eHeiden, T. L., Lloyd, D. G. \u0026amp; Ackland, T. R. Knee extension and flexion weakness in people with knee osteoarthritis: is antagonist cocontraction a factor? \u003cem\u003ejournal of orthopaedic \u0026amp; sports physical therapy\u003c/em\u003e\u003cstrong\u003e39\u003c/strong\u003e, 807-815 (2009).\u003c/li\u003e\n \u003cli\u003eWu, Y.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Weight-Bearing Physical Activity, Lower-Limb Muscle Mass, and Risk of Knee Osteoarthritis. \u003cem\u003eJAMA Network Open\u003c/em\u003e\u003cstrong\u003e7\u003c/strong\u003e, e248968-e248968 (2024).\u003c/li\u003e\n \u003cli\u003eDoorenbosch, C. A., Harlaar, J., Roebroeck, M. E. \u0026amp; Lankhorst, G. J. Two strategies of transferring from sit-to-stand; the activation of monoarticular and biarticular muscles. \u003cem\u003eJournal of biomechanics\u003c/em\u003e\u003cstrong\u003e27\u003c/strong\u003e, 1299-1307 (1994).\u003c/li\u003e\n \u003cli\u003eGismelseed, S., Al-Yahmedi, A., Zaier, R., Ouakad, H. \u0026amp; Bahadur, I. Predicting Sit-to-Stand Body Adaptation Using a Simple Model. \u003cem\u003eAxioms\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, 559 (2023).\u003c/li\u003e\n \u003cli\u003eMohajer, B.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Role of thigh muscle changes in knee osteoarthritis outcomes: osteoarthritis initiative data. \u003cem\u003eRadiology\u003c/em\u003e\u003cstrong\u003e305\u003c/strong\u003e, 169-178 (2022).\u003c/li\u003e\n \u003cli\u003ePedroso, M. G., de Almeida, A. C., Aily, J. B., de Noronha, M. \u0026amp; Mattiello, S. M. Fatty infiltration in the thigh muscles in knee osteoarthritis: a systematic review and meta-analysis. \u003cem\u003eRheumatology international\u003c/em\u003e\u003cstrong\u003e39\u003c/strong\u003e, 627-635 (2019).\u003c/li\u003e\n \u003cli\u003eKumar, D.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Quadriceps intramuscular fat fraction rather than muscle size is associated with knee osteoarthritis. \u003cem\u003eOsteoarthritis and cartilage\u003c/em\u003e\u003cstrong\u003e22\u003c/strong\u003e, 226-234 (2014).\u003c/li\u003e\n \u003cli\u003eVan Der Kruk, E. \u0026amp; Geijtenbeek, T. Is increased trunk flexion in standing up related to muscle weakness or pain avoidance in individuals with unilateral knee pain; a simulation study. \u003cem\u003eFrontiers in Bioengineering and Biotechnology\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, 1346365 (2024).\u003c/li\u003e\n \u003cli\u003eKeilstrup Ingwersen, C., Bjorholm Dahl, A., N\u0026oslash;rtoft Jensen, J. \u0026amp; Rieger Hannemose, M. Two Views Are Better than One: Monocular 3D Pose Estimation with Multiview Consistency. \u003cem\u003earXiv e-prints\u003c/em\u003e, arXiv: 2311.12421 (2023).\u003c/li\u003e\n \u003cli\u003eZhu, W.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e in \u003cem\u003eProceedings of the IEEE/CVF International Conference on Computer Vision.\u003c/em\u003e 15085-15099.\u003c/li\u003e\n \u003cli\u003eBazarevsky, V.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e Blazepose: On-device real-time body pose tracking. \u003cem\u003earXiv preprint arXiv:2006.10204\u003c/em\u003e (2020).\u003c/li\u003e\n \u003cli\u003eNair, V. \u0026amp; Hinton, G. \u003cem\u003eRectified Linear Units Improve Restricted Boltzmann Machines Vinod Nair\u003c/em\u003e. Vol. 27 (2010).\u003c/li\u003e\n \u003cli\u003eGraves, A. \u0026amp; Graves, A. Long short-term memory. \u003cem\u003eSupervised sequence labelling with recurrent neural networks\u003c/em\u003e, 37-45 (2012).\u003c/li\u003e\n \u003cli\u003eVinayavekhin, P.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e in \u003cem\u003e2018 24th International Conference on Pattern Recognition (ICPR).\u003c/em\u003e 2624-2629 (IEEE).\u003c/li\u003e\n \u003cli\u003eDai, Y., Gieseke, F., Oehmcke, S., Wu, Y. \u0026amp; Barnard, K. in \u003cem\u003eProceedings of the IEEE/CVF winter conference on applications of computer vision.\u003c/em\u003e 3560-3569.\u003c/li\u003e\n \u003cli\u003eKingma, D. P. \u0026amp; Ba, J. Adam: A method for stochastic optimization. \u003cem\u003earXiv preprint arXiv:1412.6980\u003c/em\u003e (2014).\u003c/li\u003e\n \u003cli\u003eOgier, A. C., Hostin, M.-A., Bellemare, M.-E. \u0026amp; Bendahan, D. Overview of MR image segmentation strategies in neuromuscular disorders. \u003cem\u003eFrontiers in Neurology\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, 625308 (2021).\u003c/li\u003e\n \u003cli\u003eRunge, J. in \u003cem\u003eConference on Uncertainty in Artificial Intelligence.\u003c/em\u003e 1388-1397 (Pmlr).\u003c/li\u003e\n \u003cli\u003eRunge, J.\u003cem\u003e\u0026nbsp;et al.\u003c/em\u003e (2019).\u003c/li\u003e\n \u003cli\u003eRunge, J., Gerhardus, A., Varando, G., Eyring, V. \u0026amp; Camps-Valls, G. Causal inference for time series. \u003cem\u003eNature Reviews Earth \u0026amp; Environment\u003c/em\u003e\u003cstrong\u003e4\u003c/strong\u003e, 487-505 (2023).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6225566/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6225566/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground and Objective:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eKnee osteoarthritis (OA) is a debilitating condition that compromises mobility and exacerbates knee pain, necessitating accurate and accessible diagnostic tools. Traditional motion capture technology, while effective, is often cost-prohibitive and limited to laboratory settings. In response, we developed a novel, smartphone-based approach utilizing spatiotemporal analysis of joint angular velocities and angles in sit-to-stand (STS) motion to detect symptomatic knee OA. Our deep learning model, STS-Dynamics Net, analyzed 864 sit-to-stand motion videos from 120 participants, providing a nuanced assessment of joint dynamics and temporal interactions in trunk, knee, and ankle angles and velocities. Notably, our findings demonstrate that joint angular velocities are a robust spatiotemporal biomarker for knee OA detection, outperforming the WOMAC questionnaire and maximum trunk angle in diagnostic accuracy and rivalling the performance of gold-standard 3D marker-based systems. Furthermore, our analysis revealed a significant correlation between angular velocities and muscle volumes and fat-to-muscle ratios in the quadriceps and hamstrings, underscoring the role of muscle weakness in knee OA pathogenesis. This innovative approach has the potential to revolutionize knee OA detection, enabling reliable, cost-effective, and self-administered assessments in community settings and bridging the gap in accessible healthcare monitoring.\u003c/p\u003e","manuscriptTitle":"Smartphone-Derived Joint Angular Velocities in Sit-to-Stand Motion: A Novel Spatiotemporal Marker for Symptomatic Knee Osteoarthritis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-09 15:04:49","doi":"10.21203/rs.3.rs-6225566/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"communications-medicine","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"commsmed","sideBox":"Learn more about [Communications Medicine](http://www.nature.com/commsmed)","snPcode":"43856","submissionUrl":"https://mts-commsmed.nature.com/cgi-bin/main.plex","title":"Communications Medicine","twitterHandle":"@commsmedicine","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Communications Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"051295dc-7703-4411-a2c0-60c718d61df7","owner":[],"postedDate":"April 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":46827568,"name":"Health sciences/Rheumatology/Rheumatic diseases/Osteoarthritis"},{"id":46827569,"name":"Health sciences/Biomarkers/Diagnostic markers"}],"tags":[],"updatedAt":"2026-03-09T11:13:00+00:00","versionOfRecord":[],"versionCreatedAt":"2025-04-09 15:04:49","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6225566","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6225566","identity":"rs-6225566","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}