A Deep Learning-Based Approach for Foot Placement Prediction

Human walking is an important and intuitive operation in daily life, but it is different for people missing this ability. Recognizing human gait cycle portions during walking is very useful for understanding the biomechanics of the muscles, pre-disease’s diagnosis, and designing lower limb prosthetics. The normal gait cycle is divided into stance and swing phases. In this work, benchmarking of several classification techniques is performed, based on electromyography (EMG) data collected from seven lower limb muscles, with the gait cycle phase used as a target vector to label the EMG data. The dataset is split into a training set used to build a statistical model using a specified classification technique, and a test set for purpose of testing a pre-generated statistical model. EMG signals are unfortunately normally corrupted with noise, which drastically reduces their classification performance. A median filter and root mean square (RMS) filter were thus applied to the raw EMG signals, and the performance of the classification techniques calculated in all cases for the raw EMG, EMG with median filter, and EMG with RMS filters for comparison purposes. The use of filters offered good enhancement in the classification process. The median filtered EMG signal performed best and gave higher accuracy than the raw EMG and RMS filtered EMG signals due to the technique’s efficacy in removing outliers. This work also offers a simple explanation of each classifier algorithm used, and a graphical feature selection was also applied to all seven muscles of interest to identify the muscle with the most influence on the gait cycle. The Rectus Femoris muscle shows the best activity separation between swing and stance phases, working mainly during the swing phase; this higher activity during the swing phase thus caused it to be identified as the muscle with the most influence over the normal gait cycle.

Remote sensing is a key tool to derive glacier surface velocities but existing mapping methods, such as cross-correlation techniques, can fail where surface properties change temporally or where large velocity variations occur spatially. High-resolution datasets, such as UAV imagery, offer a promising solution to tackle these issues and to study small-scale glacier dynamics, but new workflows are required to handle such data. Therefore, we tested the potential of new deep learning-based image-matching algorithms for deriving glacier surface velocities across the ablation area of a glacier with strong spatial variability in surface velocities (<5 m/yr to >100 m/yr) and substantial changes in surface properties between image acquisitions. For a thorough comparison of state-of-the-art methods and sensors, we applied three different techniques (cross-correlation using geoCosiCorr3D, feature tracking with ORB using SeaIceDrift and the new deep learning-based method using ICEpy4D) and three different platforms (Sentinel-2, PlanetScope, UAVs) to estimate glacier surface velocities. Results showed lowest errors for velocities derived with the deep learning-based approach applied to UAV imagery (RMSE = 2.17 m/yr, R2 = 0.99), followed by cross-correlation using Sentinel-2 images (RMSE = 21.0 m/yr, R2 = 0.59) and the deep learning-based approach with PlanetScope data (RMSE = 21.28 m/yr, R2 = 0.36). Cross-correlation with geoCosiCorr3D resulted in comparably high errors with the UAV dataset (RMSE = 36.22 m/yr, R2 = 0.24), whereas ORB-based feature tacking showed lowest performance with all sensors. Spatial patterns of computed velocities indicate that applying existing cross-correlation methods for areas with regular displacements or low glacier velocities yields suitable results on UAV data, but innovative deep learning-based approaches are required for resolving rapid changes in velocities or in surface properties. This novel method benefits from improved keypoint detection and matching through training using neural networks and data characterized by challenging geometries, outlier minimization and more robust descriptors by applying cross-attention layers. We conclude that continued development of deep learning-based feature tracking approaches for glacier velocity computations may substantially improve UAV-based velocity derivations applied to challenging situations. This method is able to deliver reliable displacement data in situations where traditional methods fail, which implies a new level of detail in understanding and interpreting glacier dynamics.

Kinematic Features of Rear-Foot Motion Using Anterior and Posterior Ankle-Foot Orthoses in Stroke Patients With Hemiplegic Gait

This study aims to detect gait events using a smartphone combined with deep learning and evaluate the remote effects and clinical significance of this method in different elderly populations and patients with cerebral small vessel disease (CSVD). In total, 150 healthy individuals aged 20-70 years were asked to attach a smartphone to their thighs and walk six gait cycles at self-selected low, normal, and high speeds, using an insole pressure sensor as the reference standard for gait events. A deep learning model was then established using BiTCN-BiGRU-CrossAttention, and two models (TCN-GRU and BiTCN-BiGRU) were compared. In total, 48 elderly (25 healthy, 12 with mild cognitive impairment, 11 with Parkinson's disease) participated in an online home assessment, completing single-task and cognitive dual-task walking. Overall, 35 CSVD patients participated in an offline clinical assessment, completing single-task, cognitive dual-task, and physical dual-task walking. The BiTCN-BiGRU-CrossAttention model had the lowest MAE for detecting gait events compared to the other models. All models had lower MAEs for detecting heel strikes than toe-offs, and the MAE for low and high walking was higher than for normal speed walking. There were significant differences (p < 0.05) in gait parameters (Cadence, Stride time, Stance phase, Swing phase, Stance time, Swing time, Stride length, and walking speed) between single-task and cognitive dual-task walking for all online elderly participants. CSVD patients showed significant differences (p < 0.05) in gait parameters (Cadence, Stride time, Stance phase, Swing phase, Stance time, Stride length, and walking speed) between single-task and cognitive dual-task and between single-task and physical dual-task walking.

To establish a deep learning (DL) model for predicting tumor grades and expression of pathologic markers of meningioma. A total of 1192 meningioma patients from two centers who underwent surgical resection between September 2018 and December 2021 were retrospectively included. The pathological data and post-contrast T1-weight images for each patient were collected. The patients from institute I were subdivided into training, validation, and testing sets, while the patients from institute II served as the external testing cohort. The fine-tuned ResNet50 model based on transfer learning was adopted to classify WHO grade in the whole cohort and predict Ki-67 index, H3K27me3, and progesterone receptor (PR) status of grade 1 meningiomas. The predictive performance was evaluated by the accuracy and loss curve, confusion matrix, receiver operating characteristic curve (ROC), and area under curve (AUC). The DL prediction model for each label achieved high predictive performance in two cohorts. For WHO grade prediction, the area under the curve (AUC) was 0.966 (95%CI 0.957-0.975) in the internal testing set and 0.669 (95%CI 0.643-0.695) in the external validation cohort. The AUC in predicting Ki-67 index, H3K27me3, and PR status were 0.905 (95%CI 0.895-0.915), 0.773 (95%CI 0.760-0.786), and 0.771 (95%CI 0.750-0.792) in the internal testing set and 0.591 (95%CI 0.562-0.620), 0.658 (95%CI 0.648-0.668), and 0.703 (95%CI 0.674-0.732) in the external validation cohort, respectively. DL models can preoperatively predict meningioma grades and pathologic marker expression with favorable predictive performance. Our DL model could predict meningioma grades and expression of pathologic markers and identify high-risk patients with WHO grade 1 meningioma, which would suggest a more aggressive operative intervention preoperatively and a more frequent follow-up schedule postoperatively. WHO grades and some pathologic markers of meningioma were associated with therapeutic strategies and clinical outcomes. A deep learning-based approach was employed to develop a model for predicting meningioma grades and the expression of pathologic markers. Preoperative prediction of meningioma grades and the expression of pathologic markers was beneficial for clinical decision-making.

The biomechanical effects of limb length discrepancy on gait

Monitoring the CO2 plume evolution is essential for ensuring geologic storage security and integrity. Traditional numerical simulation-based data assimilation workflow is computationally expensive, and this is further complicated by the fact that geologic uncertainty must be incorporated for robust performance prediction. Therefore, reservoir simulation and model calibration accounting for geologic uncertainty are not amenable to real time monitoring of the CO2 plume evolution for large-scale applications. We propose a deep learning-based approach which enables near real time CO2 plume visualization and rapid data assimilation incorporating multiple geological realizations for predicting future CO2 plume evolution and area of review (AOR) determination. The proposed deep learning-based data assimilation framework considers geological uncertainty utilizing multiple plausible models for training data generation. Rather than utilizing all available geologic realizations, a representative subset is sampled using dissimilarity measures in flow patterns computed via multidimensional scaling (MDS) and streamline time-of-flight. The approach substantially reduces training data generation cost while preserving the uncertainty inherent in the original ensemble of geomodels. The CO2 plume evolution is represented using ‘onset time’ images, depicting the calendar time when the CO2 saturation exceeds a prespecified threshold value at a given location. The use of a single CO2 onset time image instead of multiple CO2 saturation snapshots across different timesteps significantly reduces the dimensionality of the problem, making the deep learning model robust and scalable for large-scale field applications. A variational autoencoder encodes the onset time images into latent variables, which are predicted by another neural networks using the available monitoring data. The power and efficacy of the proposed method are demonstrated through application to a large-scale field case, the Illinois Basin-Decatur Carbon Sequestration Project. The available monitoring data consists of bottom-hole pressure at the injector, distributed pressure data and CO2 saturation log data at the monitoring well. Out of 200 geostatistical realizations, 10 representative models are selected by the MDS while preserving diversity of the geologic model. Additional calibration parameters including transmissibility and pore volume multipliers are applied to the selected realizations for generating a comprehensive training dataset. The trained ML model is then employed for reservoir model calibration, significantly accelerating the calibration process and enabling real time CO2 plume imaging from the monitoring data. The trained deep learning model achieves history matching of both pressure and saturation responses in seconds. The calibrated models are then used for forecasting future CO2 plume migration and the AOR assessment. The deep learning-based data assimilation approach enables near real time monitoring and verification of field-scale CO2 sequestration projects while accounting for geologic uncertainty. Utilizing the trained deep learning model, reservoir model calibration and prediction of CO2 plume evolution is performed within seconds, orders of magnitude faster compared to traditional history matching.

BGNN4VD: Constructing Bidirectional Graph Neural-Network for Vulnerability Detection

Ankle foot orthoses are mainly applied to provide stability in the stance phase and adequate foot clearance in the swing phase; however, they do not sufficiently assist during the entire gait cycle. On the other hand, robotic-controlled orthoses can provide mechanical assistance throughout the phases of the gait cycle. This study investigated the effect of ankle control throughout the gait cycle using an ankle joint walking assistive device under five different robotic assistance conditions: uncontrolled, dorsiflexion, and plantar flexion controlled at high and low speeds in the initial loading phase. Compared with the no-control condition, the plantar flexion condition enhanced knee extension and delayed the timing of ankle dorsiflexion in the stance phase; however, the opposite effect occurred under the dorsiflexion condition. Significant differences in the trailing limb angle and minimum toe clearance were also observed, although the same assistance was applied from the mid-stance phase to the initial swing phase. Ankle assistance in the initial loading phase affected the knee extension and ankle dorsiflexion angle during the stance phase. The smooth weight shift obtained might have a positive effect on lifting the limb during the swing phase. Robotic ankle control may provide appropriate assistance throughout the gait cycle according to individual gait ability.

BACKGROUND: Resistance exercise (RE) has been demonstrated as a superior modality for increasing muscle strength, muscle endurance, power, and motor performance. The sled used in this study is a novel device that provides proportional increase in resistance with increased speed. PURPOSE: To examine the impact of resistance on gait temporospatial parameters using a resistance sled in healthy young adults while walking and running. METHODS: Fifteen young adults (ages 21-35) were recruited to participate in this study. Sensors (accelerometers and gyroscopes) were placed on each subject at the chest, waist, both wrists, and ankles. Each participant performed three trials of 40 feet for the following conditions: self-paced walking (W), self-paced walking while pushing the sled (WP), and maximal speed running while pushing the sled (RP). RESULTS: A repeated measures MANOVA was conducted to compare gait temporospatial parameters across conditions. Results indicate significant differences (P &lt;0.005) between all conditions for stride length, cadence, double support time, swing %, and stance %. Stride length decreased across all conditions: W (85+/-3.0), WP (68+/-4.1), and RP (56+/-7.0). Cadence decreased while WP (92+/-10.1), yet increased during RP (169+/-14.9), compared to W (109+/-6.7). During WP and RP, participants demonstrated greater gait cycle percentage in stance phase [(WP, stance phase: 66+/-1.6, swing phase: 34+/-1.6) and (RP, stance phase: 57+/-2.7, swing phase: 42+/-2.7)] when compared to W (stance phase: 37+/-2.1, swing phase: 37+/-2.2). CONCLUSIONS: Longer stance phase with proportional increase in resistance could be utilized as a combined resistance and gait training tool as opposed to only gait training. Future studies should focus on neuromuscular activation of the lower extremity, specifically the muscles involved in the gait cycle stance phase, when walking or running with resistance.

Purpose - walking gait analysis is a critical field within biomechanics, providing essential insights into human motion for applications in medical diagnostics, rehabilitation, sports science, and robotics. This study presents a comprehensive review of gait analysis techniques, examining their applications, strengths, and limitations. Method - The gait cycle, comprising stance and swing phases, is analyzed using methods such as visual gait analysis, kinematic and kinetic analysis, electromyography (EMG), motion capture systems, and wearable sensors. Visual gait analysis, despite its qualitative nature, is enhanced by quantitative approaches that offer objective data. Kinematic analysis tracks limb movements, while kinetic analysis measures forces exerted during gait. EMG reveals muscle activity, aiding in neuromuscular diagnosis. Advanced technologies like motion capture and wearable sensors provide detailed and portable monitoring capabilities. Result - Integrating these methods offers a holistic view of gait, essential for clinical interventions, athletic performance enhancement, and the development of assistive technologies. Conclusion - This study underscores the importance of multi-faceted gait analysis for advancing understanding and treatment of gait abnormalities.

Segment-interaction and its relevance to the control of movement during sprinting

Human beings regularly walk over even and uneven surfaces during their daily activities. A human being with lower limb disability needs an exoskeleton to walk independently. However, walking surface irregularities increase the risk of falling of exoskeleton users. This falling tendency can be minimized by balancing the exoskeleton on irregular surface profiles against the gait cycle variation. Gait variation is studied using quality EMG signals obtained from the gastrocnemius and hamstring muscle activity during uneven surface walking. The present study compares the activity of hamstring and gastrocnemius muscles during walking on a treadmill, utilizing both even and uneven planes. Integrated electromyography signals from eight healthy male subjects are collected while walking on a treadmill, even and uneven planes. Muscle activity variation on these planes is studied using two-way ANOVA with replications. The results show that hamstring muscle activity registers a sound variation in swing phase but has no variation in stance phase over all three planes, whereas gastrocnemius muscle activity changes between swing and stance phases over even and uneven planes during forward walking. The results illustrate that the gait cycle variation depends on surface irregularities which indicates the importance of surface consideration.

The main objective of this work is to make an algorithm to detect onset and offset of stance and swing phase of human gait cycle. In order to proceed, the signal from the force sensors (placed underneath each foot) were first divided according to the fundamental period by auto-correlation. This ensures that each period contains a full stance phase and two fragmented swing phases. After this step, Short Time Fourier transform has been performed on each period of the signal to detect the initial search point (highest DC value) on the stance phase. Then by judging the width and slope of the straight line from changing search points on the signal to a changing destination point (starting from the minimum or maximum time value) of the graph the onset and offset of the stance and swing phase has been detected. The changing search point has small distance from the previous one, if the slope of the signal is small, but the distance increases with the increasing change of slope of the signal. The destination point changes with the change of slope of the straight line. This algorithm has been applied to the signals in the database of Parkinsons disease, which shows a success rate of more than 90%.

Missing data in longitudinal cognitive assessments may occur due to logistical (e.g., scheduling conflicts), personal (e.g., reduced motivation), and health-related issues (e.g., deteriorated physical or cognitive functions, anxiety, etc.). This issue poses significant challenges for clinical research and patient monitoring. Deep learning-based imputation approaches, which are agonistic to assumptions of data distribution and covariance structure, outperform traditional methods in imputing longitudinal missing data in other domains. We aim to evaluate their effectiveness in the domain of Alzheimer's Disease (AD), an area that has not been extensively explored. We developed a deep learning imputation approach using the self-attention-based imputation for time series (SAITS) model and de-identified data accessed on the AD Data Initiative's AD workbench. SAITS imputes missing values in multivariate time series by leveraging self-attention mechanisms to capture both temporal dependencies and feature correlations. It is trained by jointly optimizing imputation and reconstruction tasks on training data. We trained and evaluated SAITS using the Mini-Mental State Examination (MMSE) and Alzheimer's Disease Assessment Scale-Cognitive (ADAS-cog) test scores from GERAS-EU and GERAS-US studies, collected every 6 months over 3 years (7 visits in total). We compared SAITS' performance with (1) two state-of-the-art deep learning approaches: iTransformer and BRITS and (2) two conventional approaches: mean (averaging data from other visits of the same participant) and last observation carried forward (LOCF). We randomly split the full dataset into 80:20 training:test datasets. Deep learning models were optimized on the training set using 5-fold cross-validation and evaluated on the test set with 10% of values randomly masked/removed. Model performance was evaluated using mean absolute error (MAE) across 100 trials. Data from 1336 GERAS-EU participants (age: 72.0±11.6, 61.3% female) and 563 GERAS-US participants (age: 71.0±7.8, 52.7% female) were analyzed. SAITS performed best in most imputation tasks (Tables 1 and 2), achieving 1.636 (GERAS-EU) and 1.628 (GERAS-US) MAEs when imputing ADAS-cog total scores and 0.567 (GERAS-EU) and 1.178 (GERAS-US) MAEs when imputing MMSE total scores. Deep learning techniques demonstrates great potential in imputing missed longitudinal cognitive data in AD. Future work will evaluate their effects on down-stream tasks such as predicting cognitive decline.