Using Machine Learning Algorithms for Identifying Gait Parameters Suitable to Evaluate Subtle Changes in Gait in People with Multiple Sclerosis

Consensus quality indicators for monitoring multiple sclerosis

Multiple sclerosis (MS) as a chronic, degenerative autoimmune disease of the central nervous system has a longitudinal and heterogeneous course with increasing treatment options and risk profiles requiring constant monitoring of a growing number of parameters. Despite treatment guidelines, there is a lack of strategic and individualised monitoring pathways, including respective quality indicators (QIs). To address this, we systematically developed transparent, traceable, and measurable QIs for MS monitoring. Through literature review, expert discussions, and consensus-building, existing QIs were identified and refined. In a two-stage online Delphi process involving MS specialists (on average 53 years old and with 25 years of professional experience), the QIs were evaluated for content, clarity, and intelligibility, resulting in a set of 24 QIs and checklists to assess the quality of care. The final QIs provide a structured approach to document, monitor, and enhance the quality of care for people with MS across their treatment journey.

Torsobarography: Intra-Observer Reliability Study of a Novel Posture Analysis Based on Pressure Distribution

Postural deformities often manifest themselves in a sagittal imbalance and an asymmetric morphology of the torso. As a novel topographic method, torsobarography assesses the morphology of the back by analysing pressure distribution along the torso in a lying position. At torsobarography's core is a capacitive pressure sensor array. To evaluate its feasibility as a diagnostic tool, the reproducibility of the system and extracted anatomical associated parameters were evaluated on 40 subjects. Landmarks and reference distances were identified within the pressure images. The examined parameters describe the shape of the spine, various structures of the trunk symmetry, such as the scapulae, and the pelvic posture. The results showed that the localisation of the different structures performs with a good (ICC > 0.75) to excellent (ICC > 0.90) reliability. In particular, parameters for approximating the sagittal spine shape were reliably reproduced (ICC > 0.83). Lower reliability was observed for asymmetry parameters, which can be related to the low variability within the subject group. Nonetheless, the reliability levels of selected parameters are comparable to commercial systems. This study demonstrates the substantial potential of torsobarography at its current stage for reliable posture analysis and may pave the way as an early detection system for postural deformities.

NONAN GaitPrint: An IMU gait database of healthy young adults

An ongoing thrust of research focused on human gait pertains to identifying individuals based on gait patterns. However, no existing gait database supports modeling efforts to assess gait patterns unique to individuals. Hence, we introduce the Nonlinear Analysis Core (NONAN) GaitPrint database containing whole body kinematics and foot placement during self-paced overground walking on a 200-meter looping indoor track. Noraxon Ultium MotionTM inertial measurement unit (IMU) sensors sampled the motion of 35 healthy young adults (19-35 years old; 18 men and 17 women; mean ± 1 s.d. age: 24.6 ± 2.7 years; height: 1.73 ± 0.78 m; body mass: 72.44 ± 15.04 kg) over 18 4-min trials across two days. Continuous variables include acceleration, velocity, position, and the acceleration, velocity, position, orientation, and rotational velocity of each corresponding body segment, and the angle of each respective joint. The discrete variables include an exhaustive set of gait parameters derived from the spatiotemporal dynamics of foot placement. We technically validate our data using continuous relative phase, Lyapunov exponent, and Hurst exponent-nonlinear metrics quantifying different aspects of healthy human gait.

Using Lower Limb Wearable Sensors to Identify Gait Modalities: A Machine-Learning-Based Approach

Real-world gait analysis can aid in clinical assessments and influence related interventions, free from the restrictions of a laboratory setting. Using individual accelerometers, we aimed to use a simple machine learning method to quantify the performance of the discrimination between three self-selected cyclical locomotion types using accelerometers placed at frequently referenced attachment locations. Thirty-five participants walked along a 10 m walkway at three different speeds. Triaxial accelerometers were attached to the sacrum, thighs and shanks. Slabs of magnitude, three-second-long accelerometer data were transformed into two-dimensional Fourier spectra. Principal component analysis was undertaken for data reduction and feature selection, followed by discriminant function analysis for classification. Accuracy was quantified by calculating scalar accounting for the distances between the three centroids and the scatter of each category's cloud. The algorithm could successfully discriminate between gait modalities with 91% accuracy at the sacrum, 90% at the shanks and 87% at the thighs. Modalities were discriminated with high accuracy in all three sensor locations, where the most accurate location was the sacrum. Future research will focus on optimising the data processing of information from sensor locations that are advantageous for practical reasons, e.g., shank for prosthetic and orthotic devices.

Building a monitoring matrix for the management of multiple sclerosis

Multiple sclerosis (MS) has a longitudinal and heterogeneous course, with an increasing number of therapy options and associated risk profiles, leading to a constant increase in the number of parameters to be monitored. Even though important clinical and subclinical data are being generated, treating neurologists may not always be able to use them adequately for MS management. In contrast to the monitoring of other diseases in different medical fields, no target-based approach for a standardized monitoring in MS has been established yet. Therefore, there is an urgent need for a standardized and structured monitoring as part of MS management that is adaptive, individualized, agile, and multimodal-integrative. We discuss the development of an MS monitoring matrix which can help facilitate data collection over time from different dimensions and perspectives to optimize the treatment of people with MS (pwMS). In doing so, we show how different measurement tools can combined to enhance MS treatment. We propose to apply the concept of patient pathways to disease and intervention monitoring, not losing track of their interrelation. We also discuss the use of artificial intelligence (AI) to improve the quality of processes, outcomes, and patient safety, as well as personalized and patient-centered care. Patient pathways allow us to track the patient's journey over time and can always change (e.g., when there is a switch in therapy). They therefore may assist us in the continuous improvement of monitoring in an iterative process. Improving the monitoring process means improving the care of pwMS.

3D Printing of Individual Running Insoles - A Case Study

Purpose

The study's starting point is to find a low-cost and best-fit solution for comfortable movement for a recreational runner with knee pain using an orthopedic device. It is a case study. The research aims to apply digitization, CAD/CAM tools, and 3D printing to create an individual 3D running insole. The objective is to incorporate flexible shape optimization would provide comfort reductions in foot plantar pressures in one subject with knee pain while running. The test hypothesis was if it is possible to make it from one material. For this purpose, we created a new digital workflow based on the Decision Tree method and analyzed pain and comfort scores during user testing of prototypes.

Patient and Methods

The input data were obtained during a professional examination by a specialist doctor in the orthopedic outpatient clinic in the motion laboratory (DIERS 4D Motion Lab, Germany) with the output of data on the proband's complex movement stereotype. Surface and volumetric data were obtained in the biomedical laboratory with the 3D scanner. We modified the digital 3D foot models in 3D mesh software, developed the design in SW Gensole (Gyrobot, UK), and finally incorporated the internal structure and the surface layer of the insole data of the knowledge from the medical examination, comfort analyses, and scientific studies findings.

Results

Four complete 3D-printed prototypes (n=4) with differences in density and correction elements were designed. All of them were fabricated on a 3D printer (Prusa i3 MK3S, Czech Republic) with flexible TPU material suitable for skin contact. The Participant tested each of them five times in the field during a workout and final insoles three months on the routine training.

Conclusion

A novel workflow was created for designing, producing, and testing full 3D-printed insoles. The product is fit for immediate use.

A Pilot Study of Plantar Mechanics Distributions and Fatigue Profiles after Running on a Treadmill: Using a Support Vector Machine Algorithm

The treadmill is widely used in running fatigue experiments, and the variation of plantar mechanical parameters caused by fatigue and gender, as well as the prediction of fatigue curves by a machine learning algorithm, play an important role in providing different training programs. This experiment aimed to compare changes in peak pressure (PP), peak force (PF), plantar impulse (PI), and gender differences of novice runners after they were fatigued by running. A support vector machine (SVM) was used to predict the fatigue curve according to the changes in PP, PF, and PI before and after fatigue. 15 healthy males and 15 healthy females completed two runs at a speed of 3.3 m/s ± 5% on a footscan pressure plate before and after fatigue. After fatigue, PP, PF, and PI decreased at hallux (T1) and second-fifth toes (T2-5), while heel medial (HM) and heel lateral (HL) increased. In addition, PP and PI also increased at the first metatarsal (M1). PP, PF, and PI at T1 and T2-5 were significantly higher in females than in males, and metatarsal 3-5 (M3-5) were significantly lower in females than in males. The SVM classification algorithm results showed the accuracy was above average level using the T1 PP/HL PF (train accuracy: 65%; test accuracy: 75%), T1 PF/HL PF (train accuracy: 67.5%; test accuracy: 65%), and HL PF/T1 PI (train accuracy: 67.5%; test accuracy: 70%). These values could provide information about running and gender-related injuries, such as metatarsal stress fractures and hallux valgus. Application of the SVM to the identification of plantar mechanical features before and after fatigue. The features of the plantar zones after fatigue can be identified and the learned algorithm of plantar zone combinations with above-average accuracy (T1 PP/HL PF, T1 PF/HL PF, and HL PF/T1 PI) can be used to predict running fatigue and supervise training. It provided an important idea for the detection of fatigue after running.

Machine learning corroborates subjective ratings of walking and balance difficulty in multiple sclerosis

Machine learning can discern meaningful information from large datasets. Applying machine learning techniques to raw sensor data from instrumented walkways could automatically detect subtle changes in walking and balance. Multiple sclerosis (MS) is a neurological disorder in which patients report varying degrees of walking and balance disruption. This study aimed to determine whether machine learning applied to walkway sensor data could classify severity of self-reported symptoms in MS patients. Ambulatory people with MS (n = 107) were asked to rate the severity of their walking and balance difficulties, from 1-No problems to 5-Extreme problems, using the MS-Impact Scale-29. Those who scored less than 3 (moderately) were assigned to the “mild” group (n = 35), and those scoring higher were in the “moderate” group (n = 72). Three machine learning algorithms were applied to classify the “mild” group from the “moderate” group. The classification achieved 78% accuracy, a precision of 85%, a recall of 90%, and an F1 score of 87% for distinguishing those people reporting mild from moderate walking and balance difficulty. This study demonstrates that machine learning models can reliably be applied to instrumented walkway data and distinguish severity of self-reported impairment in people with MS.

Efficacy of Transcranial Direct Current Stimulation (tDCS) on Balance and Gait in Multiple Sclerosis Patients: A Machine Learning Approach

Transcranial direct current stimulation (tDCS) has emerged as an appealing rehabilitative approach to improve brain function, with promising data on gait and balance in people with multiple sclerosis (MS). However, single variable weights have not yet been adequately assessed. Hence, the aim of this pilot randomized controlled trial was to evaluate the tDCS effects on balance and gait in patients with MS through a machine learning approach. In this pilot randomized controlled trial (RCT), we included people with relapsing−remitting MS and an Expanded Disability Status Scale >1 and <5 that were randomly allocated to two groups—a study group, undergoing a 10-session anodal motor cortex tDCS, and a control group, undergoing a sham treatment. Both groups underwent a specific balance and gait rehabilitative program. We assessed as outcome measures the Berg Balance Scale (BBS), Fall Risk Index and timed up-and-go and 6-min-walking tests at baseline (T0), the end of intervention (T1) and 4 (T2) and 6 weeks after the intervention (T3) with an inertial motion unit. At each time point, we performed a multiple factor analysis through a machine learning approach to allow the analysis of the influence of the balance and gait variables, grouping the participants based on the results. Seventeen MS patients (aged 40.6 ± 14.4 years), 9 in the study group and 8 in the sham group, were included. We reported a significant repeated measures difference between groups for distances covered (6MWT (meters), p < 0.03). At T1, we showed a significant increase in distance (m) with a mean difference (MD) of 37.0 [−59.0, 17.0] (p = 0.003), and in BBS with a MD of 2.0 [−4.0, 3.0] (p = 0.03). At T2, these improvements did not seem to be significantly maintained; however, considering the machine learning analysis, the Silhouette Index of 0.34, with a low cluster overlap trend, confirmed the possible short-term effects (T2), even at 6 weeks. Therefore, this pilot RCT showed that tDCS may provide non-sustained improvements in gait and balance in MS patients. In this scenario, machine learning could suggest evidence of prolonged beneficial effects.

Machine learning classification of multiple sclerosis patients based on raw data from an instrumented walkway

Background

Using embedded sensors, instrumented walkways provide clinicians with important information regarding gait disturbances. However, because raw data are summarized into standard gait variables, there may be some salient features and patterns that are ignored. Multiple sclerosis (MS) is an inflammatory neurodegenerative disease which predominantly impacts young to middle-aged adults. People with MS may experience varying degrees of gait impairments, making it a reasonable model to test contemporary machine leaning algorithms. In this study, we employ machine learning techniques applied to raw walkway data to discern MS patients from healthy controls. We achieve this goal by constructing a range of new features which supplement standard parameters to improve machine learning model performance.

Results

Eleven variables from the standard gait feature set achieved the highest accuracy of 81%, precision of 95%, recall of 81%, and F1-score of 87%, using support vector machine (SVM). The inclusion of the novel features (toe direction, hull area, base of support area, foot length, foot width and foot area) increased classification accuracy by 7%, recall by 9%, and F1-score by 6%.

Conclusions

The use of an instrumented walkway can generate rich data that is generally unseen by clinicians and researchers. Machine learning applied to standard gait variables can discern MS patients from healthy controls with excellent accuracy. Noteworthy, classifications are made stronger by including novel gait features (toe direction, hull area, base of support area, foot length and foot area).

Automated Analysis of the Two-Minute Walk Test in Clinical Practice Using Accelerometer Data

One of the core problems for people with multiple sclerosis (pwMS) is the impairment of their ability to walk, which can be severely restrictive in everyday life. Therefore, monitoring of ambulatory function is of great importance to be able to effectively counteract disease progression. An extensive gait analysis, such as the Dresden protocol for multidimensional walking assessment, covers several facets of walking impairment including a 2-min walk test, in which the distance taken by the patient in two minutes is measured by an odometer. Using this approach, it is questionable how precise the measuring methods are at recording the distance traveled. In this project, we investigate whether the current measurement can be replaced by a digital measurement method based on accelerometers (six Opal sensors from the Mobility Lab system) that are attached to the patient’s body. We developed two algorithms using these data and compared the validity of these approaches using the results from 2-min walk tests from 562 pwMS that were collected with a gold-standard odometer. In 48.4% of pwMS, we detected an average relative measurement error of less than 5%, while results from 25.8% of the pwMS showed a relative measurement error of up to 10%. The algorithm had difficulties correctly calculating the walking distances in another 25.8% of pwMS; these results showed a measurement error of more than 20%. A main reason for this moderate performance was the variety of pathologically altered gait patterns in pwMS that may complicate the step detection. Overall, both algorithms achieved favorable levels of agreement (r = 0.884 and r = 0.980) with the odometer. Finally, we present suggestions for improvement of the measurement system to be implemented in the future.