Background and aim Regular physical activity preserves functional independence in older adults, yet care-home residents often miss out because personalized supervision is scarce. Autonomous, technology-supported exercise platforms could deliver such guidance without additional staff time—but only if sessions are automatically monitored for safety and quality. We therefore designed a deep learning (DL) system that (a) recognizes individual exercise types and (b) estimates joint angle trajectories from a standard video recording. These outputs are used to compute objective exercise performance metrics (EPMs) such as duration, repetition count, motion variability, and range of motion.

Methods Seven care-home residents (aged between 65–94 years) performed six common rehabilitation exercises in front of a single camera while wearing 17 inertial sensors (Xsens MVN Awinda) that provided ground-truth joint angles. Two-dimensional skeleton poses estimated from the video were fed into a temporal convolutional neural network to recognize the exercises and estimate three-dimensional joint angles. We evaluated exercise segmentation with F1@50 and angle regression with mean per-joint angular error (MPJAE) across nine trunk and lower-limb joints, using leave-one-subject-out cross-validation. Pearson correlations assessed agreement between estimated and ground-truth EPMs.

Results The DL model achieved an F1@50 of 0.92 (± 0.04) for exercise recognition and an MPJAE of 7.7° (± 0.91) for joint angle estimation. The estimated EPMs aligned closely with ground truth, achieving correlation scores of 0.93 (95% CI [0.90, 0.95]) for duration, 0.86 (95% CI [0.80, 0.90]) for repetition count, and between 0.3 and 0.9 for motion variability and range of motion across exercises.

Conclusion The DL algorithm reliably estimates key exercise outcomes from a single video stream. This video-based monitoring pipeline could enable unsupervised, technology-supported exercise assessment in residential care homes while safeguarding session quality and safety. Future work will validate the approach in larger and more diverse cohorts.

Freezing of gait (FOG) is a debilitating symptom of Parkinson's disease (PD), characterized by an absence or reduction in forward movement of the legs despite the intention to walk. Detecting FOG during free-living conditions presents significant challenges, particularly when using only inertial measurement unit (IMU) data, as it must be distinguished from voluntary stopping events that also feature reduced forward movement. Influences from stress and anxiety, measurable through galvanic skin response (GSR) and electrocardiogram (ECG), may assist in distinguishing FOG from normal gait and stopping. However, no study has investigated the fusion of IMU, GSR, and ECG for FOG detection. Therefore, this study introduced two methods: a two-step approach that first identified reduced forward movement segments using a Transformer-based model with IMU data, followed by an XGBoost model classifying these segments as FOG or stopping using IMU, GSR, and ECG features; and an end-to-end approach employing a multi-stage temporal convolutional network to directly classify FOG and stopping segments from IMU, GSR, and ECG data. Results showed that the two-step approach with all data modalities achieved an average F1 score of 0.728 and F1@50 of 0.725, while the end-to-end approach scored 0.771 and 0.759, respectively. However, no significant difference was found compared to using only IMU data in both approaches (p-values: 0.466 to 0.887). In conclusion, adding physiological data did not provide a statistically significant benefit in distinguishing between FOG and stopping. The limitations may be specific to GSR and ECG data, and may not generalize to other physiological modalities.

Video annotation is the gold-standard method to assess Freezing of Gait (FOG) in Parkinsonian disorders, but it is time-consuming. Deep learning (DL)-based assessment of FOG using inertial measurement units ameliorates these problems but poses challenges. Particularly, the large heterogeneity between patients and assessment methods potentially affects detection performance between independent cohorts. To evaluate heterogeneity effects, we developed a DL model on a local cohort (85 participants; 2043 trials) and validated it across six external cohorts (256 participants; 1058 trials). Model-expert agreement on the percentage-of-time-frozen was strong locally (ICC=0.886 [0.79,0.90]) but reduced in external cohorts (ICC=0.562±0.141). Fine-tuning the DL model with just 50 minutes of external cohort data improved the ICC to 0.732±0.138, falling within the borderline of the inter-rater agreement (ICC=0.73-0.99). Therefore, while unified standards are still being developed, we propose an expert-in-the-loop workflow as an effective intermediary and present a proof-of-concept web-based platform for fine-tuning and expert review (aidfog.be).

Background: Freezing of gait (FOG) is an episodic and highly disabling symptom of Parkinson’s Disease (PD). Traditionally, FOG assessment relies on time-consuming visual inspection of camera footage. Therefore, previous studies have proposed portable and automated solutions to annotate FOG. However, automated FOG assessment is challenging due to gait variability caused by medication effects and varying FOG-provoking tasks. Moreover, whether automated approaches can differentiate FOG from typical everyday movements, such as volitional stops, remains to be determined. To address these questions, we evaluated an automated FOG assessment model with deep learning (DL) based on inertial measurement units (IMUs). We assessed its performance trained on all standardized FOG-provoking tasks and medication states, as well as on specific tasks and medication states. Furthermore, we examined the effect of adding stopping periods on FOG detection performance. Methods: Twelve PD patients with self-reported FOG (mean age 69.33 ± 6.02 years) completed a FOG-provoking protocol, including timed-up-and-go and 360-degree turning-in-place tasks in On/Off dopaminergic medication states with/without volitional stopping. IMUs were attached to the pelvis and both sides of the tibia and talus. A temporal convolutional network (TCN) was used to detect FOG episodes. FOG severity was quantified by the percentage of time frozen (%TF) and the number of freezing episodes (#FOG). The agreement between the model-generated outcomes and the gold standard experts’ video annotation was assessed by the intra-class correlation coefficient (ICC). Results: For FOG assessment in trials without stopping, the agreement of our model was strong (ICC (%TF) = 0.92 [0.68, 0.98]; ICC(#FOG) = 0.95 [0.72, 0.99]). Models trained on a specific FOG-provoking task could not generalize to unseen tasks, while models trained on a specific medication state could generalize to unseen states. For assessment in trials with stopping, the agreement of our model was moderately strong (ICC (%TF) = 0.95 [0.73, 0.99]; ICC (#FOG) = 0.79 [0.46, 0.94]), but only when stopping was included in the training data. Conclusion: A TCN trained on IMU signals allows valid FOG assessment in trials with/without stops containing different medication states and FOG-provoking tasks. These results are encouraging and enable future work investigating automated FOG assessment during everyday life.

— Freezing of gait (FOG) is an episodic and highly disabling symptom of Parkinson’s disease (PD). Although described as a single phenomenon, FOG is heterogeneous and can express as different manifestations, such as trembling in place or complete akinesia. We aimed to analyze the efficacy of deep learning (DL) trained on inertial measurement unit data to classify FOG into both manifestations. We adapted and compared four state-of-the-art FOG detection algorithms for this task and investigated the advantages of incorporating a refinement model to address oversegmentation errors. We evaluated the model’s performance in distinguishing between trembling and akinesia, as well as other forms of movement cessation (e.g., stopping and sitting), against gold-standard video annotations. Experiments were conducted on a dataset of eighteen PD patients completing a FOG-provoking protocol in a gait laboratory. Results showed our model achieved an F1 score of 0.78 and segment F1@50 of 0.75 in detecting FOG manifestations. Assessment of FOG severity was strong for trembling (ICC=0.86, [0.66,0.95]) and moderately strong for akinesia (ICC=0.78, [0.51,0.91]). Importantly, our model successfully differentiated FOG from other forms of movement cessation during 360-degree turning-in-place tasks. In conclusion, our study demonstrates that DL can accurately assess different types of FOG manifestations, warranting further investigation in larger and more diverse verification cohorts.

Freezing of gait (FOG) is a common and severe symptom of Parkinson's disease (PD). Due to the complex underlying pathophysiology, FOG is difficult to assess, hampering further insight into this phenomenon. Inertial measurement units (IMUs) may enable FOG assessment during everyday life, but lack of standardization, e.g., the number and position of the IMUs, complicates an objective comparison of automatic FOG assessment algorithms. We propose a multi-stage temporal dilated convolutional model to automatically assess FOG based on IMU data. We collected simultaneous optical motion capture (MoCap) and IMU data of ten people with PD and FOG. We devised a simulation pipeline, i.e., generating IMU data from MoCap data, to objectively compare our approach to two state-of-The-Art FOG assessment models. The comparison was performed for five simulated IMU configurations, ranging from 1 to 7 IMUs. The results show that our approach outperforms the two state-of-The-Art methods on most of the simulated IMU configurations. The complete lower-body IMU setup of 7 IMUs (pelvis and both sides of the talus, tibia, and femur) enables the best FOG detection performance. Lastly, we show that our model trained by incorporating simulated IMU data enabled significantly improved FOG detection performance than our model trained only with real IMU data. In doing so, we demonstrate that retrospective MoCap datasets can be re-used to train expressive IMU-based FOG assessment models, reducing the required amount of dedicated and labor-intensive IMU data collection experiments.