Effect of sensor number and location on accelerometry-based vertical ground reaction force estimation during walking.

Knee osteoarthritis is a major cause of global disability and is a major cost for the healthcare system. Lower extremity loading is a determinant of knee osteoarthritis onset and progression; however, technology that assists rehabilitative clinicians in optimizing key metrics of lower extremity loading is significantly limited. The peak vertical component of the ground reaction force (vGRF) in the first 50% of stance is highly associated with biological and patient-reported outcomes linked to knee osteoarthritis symptoms. Monitoring and maintaining typical vGRF profiles may support healthy gait biomechanics and joint tissue loading to prevent the onset and progression of knee osteoarthritis. Yet, the optimal number of sensors and sensor placements for predicting accurate vGRF from accelerometry remains unknown. Our goals were to: 1) determine how many sensors and what sensor locations yielded the most accurate vGRF loading peak estimates during walking; and 2) characterize how prescribing different loading conditions affected vGRF loading peak estimates. We asked 20 young adult participants to wear 5 accelerometers on their waist, shanks, and feet and walk on a force-instrumented treadmill during control and targeted biofeedback conditions prompting 5% underloading and overloading vGRFs. We trained and tested machine learning models to estimate vGRF from the various sensor accelerometer inputs and identified which combinations were most accurate. We found that a neural network using one accelerometer at the waist yielded the most accurate loading peak vGRF estimates during walking, with average errors of 4.4% body weight. The waist-only configuration was able to distinguish between control and overloading conditions prescribed using biofeedback, matching measured vGRF outcomes. Including foot or shank acceleration signals in the model reduced accuracy, particularly for the overloading condition. Our results suggest that a system designed to monitor changes in walking vGRF or to deploy targeted biofeedback may only need a single accelerometer located at the waist for healthy participants.

Neurocognitive predictors of adherence to an online pain self-management program adjunct to long-term opioid therapy.

INTRODUCTION: While pain self-management programs can significantly improve patient outcomes, poor adherence is common and the need for research on predictors of adherence has been noted. A potential, but commonly overlooked, predictor is cognitive function. Our aim, then, was to examine the relative influence of various cognitive functional domains on engagement with an online pain self-management program. METHOD: A secondary analysis of a randomized controlled trial testing the impact of E-health (a 4-month subscription to the online Goalistics Chronic Pain Management Program) plus treatment as usual, relative to treatment as usual alone, on pain and opioid dose outcomes in adults receiving long-term opioid therapy of morphine equivalence dose ≥20 mg; 165 E-health participants who completed an on-line neurocognitive battery were included in this sub-analysis. A variety of demographic, clinical, and symptom rating scales were also examined. We hypothesized that better processing speed and executive functions at baseline would predict engagement with the 4-month E-health subscription. RESULTS: Ten functional cognitive domains were identified using exploratory factor analysis and the resultant factor scores applied for hypothesis testing. The strongest predictors of E-health engagement were selective attention, and response inhibition and speed domains. An explainable machine learning algorithm improved classification accuracy, sensitivity, and specificity. CONCLUSIONS: The results suggest that cognition, especially selective attention, inhibitory control, and processing speed, is predictive of online chronic pain self-management program engagement. Future research to replicate and extend these findings seems warranted. CLINICALTRIALS.GOV REGISTRATION NUMBER: NCT03309188.

Interpretable AI for bio-medical applications.

This paper presents the use of two popular explainability tools called Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive exPlanations (SHAP) to explain the predictions made by a trained deep neural network. The deep neural network used in this work is trained on the UCI Breast Cancer Wisconsin dataset. The neural network is used to classify the masses found in patients as benign or malignant based on 30 features that describe the mass. LIME and SHAP are then used to explain the individual predictions made by the trained neural network model. The explanations provide further insights into the relationship between the input features and the predictions. SHAP methodology additionally provides a more holistic view of the effect of the inputs on the output predictions. The results also present the commonalities between the insights gained using LIME and SHAP. Although this paper focuses on the use of deep neural networks trained on UCI Breast Cancer Wisconsin dataset, the methodology can be applied to other neural networks and architectures trained on other applications. The deep neural network trained in this work provides a high level of accuracy. Analyzing the model using LIME and SHAP adds the much desired benefit of providing explanations for the recommendations made by the trained model.

The Intelligent Phenotypic Plasticity Platform (IP3) for Precision Medicine-Based Injury Prevention in Sport.

The best predictor of future injury is previous injury and this has not changed in a quarter century despite the introduction of evidence-based medicine and associated revisions to post-injury treatment and care. Nearly nine million sports-related injuries occur annually, and the majority of these require medical intervention prior to clearance for the athlete to return to play (RTP). Regardless of formal care, these athletes remain two to four times more likely to suffer a second injury for several years after RTP. In the case of children and young adults, this sets them up for a lifetime of negative health outcomes. Thus, the initial injury is the tipping point for a post-injury cascade of negative sequelae exposing athletes to more physical and psychological pain, higher medical costs, and greater risk of severe long-term negative health throughout their life. This chapter details the technologies and method that make up the automated Intelligent Phenotypic Plasticity Platform (IP3)-a revolutionary new approach to the current standard of post-injury care that identifies and targets deficits that underly second injury risk in sport. IP3 capitalizes on the biological concept of phenotypic plasticity (PP) to quantify an athlete's functional adaptability across different performance environments, and it is implemented in two distinct steps: (1) phenomic profiling and (2) precision treatment. Phenomic profiling indexes the fitness and subsequent phenotypic plasticity of an individual athlete, which drives the personalization of the precision treatment step. IP3 leverages mixed-reality technologies to present true-to-life environments that test the athlete's ability to adapt to dynamic stressors. The athlete's phenotypic plasticity profile is then used to drive a precision treatment that systematically stresses the athlete, via a combination of behavioral-based and genetic fuzzy system models, to optimally enhance the athlete's functional adaptability. IP3 is computationally light-weight and, through the integration with mixed-reality technologies, promotes real-time prediction, responsiveness, and adaptation. It is also the first ever phenotypic plasticity-based precision medicine platform, and the first precision sports medicine platform of any kind.

Genetic Fuzzy Based Scalable System of Distributed Robots for a Collaborative Task.

This paper introduces a new genetic fuzzy based paradigm for developing scalable set of decentralized homogenous robots for a collaborative task. In this work, the number of robots in the team can be changed without any additional training. The dynamic problem considered in this work involves multiple stationary robots that are assigned with the goal of bringing a common effector, which is physically connected to each of these robots through cables, to any arbitrary target position within the workspace of the robots. The robots do not communicate with each other. This means that each robot has no explicit knowledge of the actions of the other robots in the team. At any instant, the robots only have information related to the common effector and the target. Genetic Fuzzy System (GFS) framework is used to train controllers for the robots to achieve the common goal. The same GFS model is shared among all robots. This way, we take advantage of the homogeneity of the robots to reduce the training parameters. This also provides the capability to scale to any team size without any additional training. This paper shows the effectiveness of this methodology by testing the system on an extensive set of cases involving teams with different number of robots. Although the robots are stationary, the GFS framework presented in this paper does not put any restriction on the placement of the robots. This paper describes the scalable GFS framework and its applicability across a wide set of cases involving a variety of team sizes and robot locations. We also show results in the case of moving targets.