The use of artificial neural networks to identify patients with chronic low-back pain conditions from patterns of sit-to-stand manoeuvres

Machine learning in the diagnosis, management, and care of patients with low back pain: a scoping review of the literature and future directions

BACKGROUND CONTEXT

Low back pain (LBP) remains the leading cause of disability globally. In recent years, machine learning (ML) has emerged as a potentially useful tool to aid the diagnosis, management, and prognostication of LBP.

PURPOSE

In this review, we assess the scope of ML applications in the LBP literature and outline gaps and opportunities.

STUDY DESIGN/SETTING

A scoping review was performed in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) guidelines.

METHODS

Articles were extracted from the Web of Science, Scopus, PubMed, and IEEE Xplore databases. Title/abstract and full-text screening was performed by two reviewers. Data on model type, model inputs, predicted outcomes, and ML methods were collected.

RESULTS

In total, 223 unique studies published between 1988 and 2023 were identified, with just over 50% focused on low-back-pain detection. Neural networks were used in 106 of these articles. Common inputs included patient history, demographics, and lab values (67% total). Articles published after 2010 were also likely to incorporate imaging data into their models (41.7% of articles). Of the 212 supervised learning articles identified, 168 (79.4%) mentioned use of a training or testing dataset, 116 (54.7%) utilized cross-validation, and 46 (21.7%) implemented hyperparameter optimization. Of all articles, only 8 included external validation and 9 had publicly available code.

CONCLUSIONS

Despite the rapid application of ML in LBP research, a majority of articles do not follow standard ML best practices. Furthermore, over 95% of articles cannot be reproduced or authenticated due to lack of code availability. Increased collaboration and code sharing are needed to support future growth and implementation of ML in the care of patients with LBP.

NSF DARE-transforming modeling in neurorehabilitation: a patient-in-the-loop framework

In 2023, the National Science Foundation (NSF) and the National Institute of Health (NIH) brought together engineers, scientists, and clinicians by sponsoring a conference on computational modelling in neurorehabiilitation. To facilitate multidisciplinary collaborations and improve patient care, in this perspective piece we identify where and how computational modelling can support neurorehabilitation. To address the where, we developed a patient-in-the-loop framework that uses multiple and/or continual measurements to update diagnostic and treatment model parameters, treatment type, and treatment prescription, with the goal of maximizing clinically-relevant functional outcomes. This patient-in-the-loop framework has several key features: (i) it includes diagnostic and treatment models, (ii) it is clinically-grounded with the International Classification of Functioning, Disability and Health (ICF) and patient involvement, (iii) it uses multiple or continual data measurements over time, and (iv) it is applicable to a range of neurological and neurodevelopmental conditions. To address the how, we identify state-of-the-art and highlight promising avenues of future research across the realms of sensorimotor adaptation, neuroplasticity, musculoskeletal, and sensory & pain computational modelling. We also discuss both the importance of and how to perform model validation, as well as challenges to overcome when implementing computational models within a clinical setting. The patient-in-the-loop approach offers a unifying framework to guide multidisciplinary collaboration between computational and clinical stakeholders in the field of neurorehabilitation.

ESTIMATION OF KNEE JOINT TORQUE DURING SIT–STAND MOVEMENT BASED ON sEMG SIGNALS USING NEURAL NETWORKS

The estimation of knee joint torque is important for the development of powered exoskeletons to achieve ideal gait characteristics. In this study, we proposed three different models to predict the required torque for performing sit-to-stand (STS) and back-to-sit (BTS) movements. The surface electromyography (sEMG) signals were extracted from the biceps femoris and rectus femoris muscles during STS and BTS movements. The time-domain features selected as input to the models for torque prediction are integrated EMG (iEMG), root mean square (RMS), and mean absolute value (MAV). Two-way ANOVA analysis identifies the significance of NN models and EMG features of the muscles in predicting the knee joint torque requirement. The artificial neural network models selected for prediction are the feed-forward back-propagation algorithm, ANFIS, and NARX. The theoretical value of knee joint torque calculated using the Lagrange method was compared with the torque output for each model based on root mean square error (RMSE). The desired torque predicted using the NARX model confirms to have the least average error ([Formula: see text][Formula: see text]Nm), which indicates that NARX can estimate knee joint torque more accurately from sEMG than other models.

Mathematical and computational models for pain: a systematic review

OBJECTIVE

There is no single prevailing theory of pain that explains its origin, qualities, and alleviation. Although many studies have investigated various molecular targets for pain management, few have attempted to examine the etiology or working mechanisms of pain through mathematical or computational model development. In this systematic review, we identified and classified mathematical and computational models for characterizing pain.

METHODS

The databases queried were Science Direct and PubMed, yielding 560 articles published prior to January 1st, 2020. After screening for inclusion of mathematical or computational models of pain, 31 articles were deemed relevant.

RESULTS

Most of the reviewed articles utilized classification algorithms to categorize pain and no-pain conditions. We found the literature heavily focused on the application of existing models or machine learning algorithms to identify the presence or absence of pain, rather than to explore features of pain that may be used for diagnostics and treatment.

CONCLUSIONS

Although understudied, the development of mathematical models may augment the current understanding of pain by providing directions for testable hypotheses of its underlying mechanisms. Additional focus is needed on developing models that seek to understand the underlying mechanisms of pain, as this could potentially lead to major breakthroughs in its treatment.

Kinematic biomarkers of chronic neck pain measured during gait: A data-driven classification approach

People with chronic neck pain (CNP) often present with altered gait kinematics. This paper investigates, combines, and compares the kinematic features from linear and nonlinear walking trajectories to design supervised machine learning models which differentiate asymptomatic individuals from those with CNP. For this, 126 features were extracted from seven body segments of 20 asymptomatic subjects and 20 individuals with non-specific CNP. Neighbourhood Component Analysis (NCA) was used to identify body segments and the corresponding significant features which have the maximum discriminative power for conducting classification. We assessed the efficacy of NCA combined with K- Nearest Neighbour (K-NN), Support Vector Machine and Linear Discriminant Analysis. By applying NCA, all classifiers increased their performance for both linear and nonlinear walking trajectories. Notably, features selected by NCA which magnify the classification power of the computational model were solely from the head, trunk and pelvis kinematics. Our results revealed that the nonlinear trajectory provides the best classification performance through the NCA-K-NN algorithms with an accuracy of 90%, specificity of 100% and sensitivity of 83.3%. The selected features by NCA are introduced as key biomarkers of gait kinematics for classifying non-specific CNP. This paper provides insight into changes in gait kinematics which are present in people with non-specific CNP which can be exploited for classification purposes. The result highlights the importance of curvilinear gait kinematic features which potentially could be utilized in future research to predict recurrent episodes of neck pain.

Does Artificial Intelligence Outperform Natural Intelligence in Interpreting Musculoskeletal Radiological Studies? A Systematic Review

BACKGROUND

Machine learning (ML) is a subdomain of artificial intelligence that enables computers to abstract patterns from data without explicit programming. A myriad of impactful ML applications already exists in orthopaedics ranging from predicting infections after surgery to diagnostic imaging. However, no systematic reviews that we know of have compared, in particular, the performance of ML models with that of clinicians in musculoskeletal imaging to provide an up-to-date summary regarding the extent of applying ML to imaging diagnoses. By doing so, this review delves into where current ML developments stand in aiding orthopaedists in assessing musculoskeletal images.

QUESTIONS/PURPOSES

This systematic review aimed (1) to compare performance of ML models versus clinicians in detecting, differentiating, or classifying orthopaedic abnormalities on imaging by (A) accuracy, sensitivity, and specificity, (B) input features (for example, plain radiographs, MRI scans, ultrasound), (C) clinician specialties, and (2) to compare the performance of clinician-aided versus unaided ML models.

METHODS

A systematic review was performed in PubMed, Embase, and the Cochrane Library for studies published up to October 1, 2019, using synonyms for machine learning and all potential orthopaedic specialties. We included all studies that compared ML models head-to-head against clinicians in the binary detection of abnormalities in musculoskeletal images. After screening 6531 studies, we ultimately included 12 studies. We conducted quality assessment using the Methodological Index for Non-randomized Studies (MINORS) checklist. All 12 studies were of comparable quality, and they all clearly included six of the eight critical appraisal items (study aim, input feature, ground truth, ML versus human comparison, performance metric, and ML model description). This justified summarizing the findings in a quantitative form by calculating the median absolute improvement of the ML models compared with clinicians for the following metrics of performance: accuracy, sensitivity, and specificity.

RESULTS

ML models provided, in aggregate, only very slight improvements in diagnostic accuracy and sensitivity compared with clinicians working alone and were on par in specificity (3% (interquartile range [IQR] -2.0% to 7.5%), 0.06% (IQR -0.03 to 0.14), and 0.00 (IQR -0.048 to 0.048), respectively). Inputs used by the ML models were plain radiographs (n = 8), MRI scans (n = 3), and ultrasound examinations (n = 1). Overall, ML models outperformed clinicians more when interpreting plain radiographs than when interpreting MRIs (17 of 34 and 3 of 16 performance comparisons, respectively). Orthopaedists and radiologists performed similarly to ML models, while ML models mostly outperformed other clinicians (outperformance in 7 of 19, 7 of 23, and 6 of 10 performance comparisons, respectively). Two studies evaluated the performance of clinicians aided and unaided by ML models; both demonstrated considerable improvements in ML-aided clinician performance by reporting a 47% decrease of misinterpretation rate (95% confidence interval [CI] 37 to 54; p < 0.001) and a mean increase in specificity of 0.048 (95% CI 0.029 to 0.068; p < 0.001) in detecting abnormalities on musculoskeletal images.

CONCLUSIONS

At present, ML models have comparable performance to clinicians in assessing musculoskeletal images. ML models may enhance the performance of clinicians as a technical supplement rather than as a replacement for clinical intelligence. Future ML-related studies should emphasize how ML models can complement clinicians, instead of determining the overall superiority of one versus the other. This can be accomplished by improving transparent reporting, diminishing bias, determining the feasibility of implantation in the clinical setting, and appropriately tempering conclusions.

LEVEL OF EVIDENCE

Level III, diagnostic study.

Sensor Technologies to Manage the Physiological Traits of Chronic Pain: A Review

Non-oncologic chronic pain is a common high-morbidity impairment worldwide and acknowledged as a condition with significant incidence on quality of life. Pain intensity is largely perceived as a subjective experience, what makes challenging its objective measurement. However, the physiological traces of pain make possible its correlation with vital signs, such as heart rate variability, skin conductance, electromyogram, etc., or health performance metrics derived from daily activity monitoring or facial expressions, which can be acquired with diverse sensor technologies and multisensory approaches. As the assessment and management of pain are essential issues for a wide range of clinical disorders and treatments, this paper reviews different sensor-based approaches applied to the objective evaluation of non-oncological chronic pain. The space of available technologies and resources aimed at pain assessment represent a diversified set of alternatives that can be exploited to address the multidimensional nature of pain.

Is there evidence to use kinematic/kinetic measures clinically in low back pain patients? A systematic review

BACKGROUND

Currently, there is a widespread reliance on self-reported questionnaires to assess low back pain patients. However, it has been suggested that objective measures of low back pain patients' functional status should be used to aid clinical assessment. The aim of this study is to systematically review which kinematic /kinetic parameters have been used to assess low back pain patients against healthy controls and to propose clinical kinematic/kinetic measures.

METHODS

PubMed, Embase and Scopus databases were searched for relevant studies. Reference lists of selected studies and hand searches were performed. Studies had to compare people with and without non-specific low back pain while performing functional tasks and report body segment/joint kinematic and/or kinetic data. Two reviewers independently identified relevant papers.

FINDINGS

Sixty-two studies were included. Common biases identified were lack of assessor blinding and sample size calculation, use of samples of convenience, and poor experimental protocol standardization. Studies had small sample sizes. Range of motion maneuvers were the main task performed (33/62). Kinematic/kinetic data of different individual or combination of body segments/joints were reported among the studies, commonest was to assess the hip joint and lumbar segment motion (13/62). Only one study described full body movement. The most commonly reported outcome was range of motion. Statistically significant differences between controls and low back pain groups were reported for different outcomes among the studies. Moreover, when the same outcome was reported disagreements were noted.

INTERPRETATION

The literature to date offers limited and inconsistent evidence of kinematic/kinetic measures in low back pain patients that could be used clinically.

Multi-factorial causative model for back pain management; relating causative factors and mechanisms to injury presentations and designing time- and cost effective treatment thereof

Back pain resolution has not statistically improved over many years with some literature suggesting chronic back pain to be increasing. From a search of literature on causes, events, mechanisms, factors and treatment for back pain, a model is developed that relates causes of back injury to factors that result in pain through two primary mechanisms; muscle fatigue and muscle/tendon/connective tissue strain or sprain with other main mechanisms being diminished reactivity and strength, changes in tendon/tissue mechanical properties and fear of back pain recurrence/fear of movement following a back pain episode. The model highlights the fact that back pain/injury is multi-factorial with numerous circular relationships. Therefore treatment should also be multi-factorial; a combination of physical and psychological therapy with attention to mechanisms at work or in daily living that exacerbate the injury and delay recovery thereof. Exercise is one method that can reduce muscle imbalance, improve resilience to muscle fatigue, and address reactivity and strength. More importantly, eccentric exercise can rectify musculotendinous or connective tissue injury which plays a role in prolonging the back injury cycle. Posture is identified as a causative factor for back pain with the time exposure for posture representing the largest portion of daily activities. From literature and from clinical observation, treatment methods can be improved and incorporated into integrated multi-modal programs. An integrated exercise program that commences with motor control exercise and progresses into functional movement is suggested. Furthermore a modification of the McKenzie extension movement may benefit back injury rehabilitation for a majority of lower back pain patients. Otherwise the sit-to-stand movement is a regular and frequent exacerbating mechanism of back pain and likely continuously tears connective tissue during the movement thus prolonging the cycle of back pain and can be addressed instantly with a modification in sit-to-stand technique.

Individuals with non-specific low back pain use a trunk stiffening strategy to maintain upright posture

There is increasing evidence that individuals with non-specific low back pain (LBP) have altered movement coordination. However, the relationship of this neuromotor impairment to recurrent pain episodes is unknown. To assess coordination while minimizing the confounding influences of pain we characterized automatic postural responses to multi-directional support surface translations in individuals with a history of LBP who were not in an active episode of their pain. Twenty subjects with and 21 subjects without non-specific LBP stood on a platform that was translated unexpectedly in 12 directions. Net joint torques of the ankles, knees, hips, and trunk in the frontal and sagittal planes as well as surface electromyographs of 12 lower leg and trunk muscles were compared across perturbation directions to determine if individuals with LBP responded using a trunk stiffening strategy. Individuals with LBP demonstrated reduced peak trunk torques, and enhanced activation of the trunk and ankle muscle responses following perturbations. These results suggest that individuals with LBP use a strategy of trunk stiffening achieved through co-activation of trunk musculature, aided by enhanced distal responses, to respond to unexpected support surface perturbations. Notably, these neuromotor alterations persisted between active pain periods and could represent either movement patterns that have developed in response to pain or could reflect underlying impairments that may contribute to recurrent episodes of LBP.

Effects of experimentally induced low back pain on the sit-to-stand movement and electroencephalographic contingent negative variation

It is becoming increasingly evident that people with chronic, recurrent low back pain (LBP) exhibit changes in cerebrocortical activity that associate with altered postural coordination, suggesting a need for a better understanding of how the experience of LBP alters postural coordination and cerebrocortical activity. To characterize changes in postural coordination and pre-movement cerebrocortical activity related to the experience of acutely induced LBP, 14 healthy participants with no history of LBP performed sit-to-stand movements in 3 sequential conditions: (1) without experimentally induced LBP; NoPain1, (2) with movement-associated LBP induced by electrocutaneous stimulation; Pain, and (3) again without induced LBP; NoPain2. The Pain condition elicited altered muscle activation and redistributed forces under the seat and feet prior to movement, decreased peak vertical force exerted under the feet during weight transfer, longer movement times, as well as decreased and earlier peak hip extension. Stepwise regression models demonstrated that electroencephalographic amplitudes of contingent negative variation during the Pain condition significantly correlated with the participants' change in sit-to-stand measures between the NoPain1 and Pain conditions, as well as with the subsequent difference in sit-to-stand measures between the NoPain1 and NoPain2 conditions. The results, therefore, identify the contingent negative variation as a correlate for the extent of an individual's LBP-related movement modifications and to the subsequent change in movement patterns from before to after the experience of acutely induced LBP, thereby providing a direction for future studies aimed to understand the neural mechanisms underlying the development of altered movement patterns with LBP.

Prolonged weight-shift and altered spinal coordination during sit-to-stand in practitioners of the Alexander Technique

The Alexander Technique (AT) is used to improve postural and movement coordination and has been reported to be clinically beneficial, however its effect on movement coordination is not well-characterized. In this study we examined the sit-to-stand (STS) movement by comparing coordination (phasing, weight-shift and spinal movement) between AT teachers (n=15) and matched control subjects (n=14). We found AT teachers had a longer weight-shift (p<0.001) and shorter momentum transfer phase (p=0.01), than control subjects. AT teachers also increased vertical foot force monotonically, rather than unweighting the feet prior to seat-off, suggesting they generate less forward momentum with hip flexors. The prolonged weight-shift of AT teachers occurred over a greater range of trunk inclination, such that their weight shifted continuously onto the feet while bringing the body mass forward. Finally, AT teachers had greatly reduced spinal bending during STS (cervical, p<0.001; thoracic, p<0.001; lumbar, p<0.05). We hypothesize that the low hip joint stiffness and adaptive axial postural tone previously reported in AT teachers underlies this novel "continuous" STS strategy by facilitating eccentric contractions during weight-shift.

Prediction of low back pain with two expert systems

Low back pain (LBP) is one of the common problems encountered in medical applications. This paper proposes two expert systems (artificial neural network and adaptive neuro-fuzzy inference system) for the assessment of the LBP level objectively. The skin resistance and visual analog scale (VAS) values have been accepted as the input variables for the developed systems. The results showed that the expert systems behave very similar to real data and that use of the expert systems can be used to successfully diagnose the back pain intensity. The suggested systems were found to be advantageous approaches in addition to existing unbiased approaches. So far as the authors are aware, this is the first attempt of using the two expert systems achieving very good performance in a real application. In light of some of the limitations of this study, we also identify and discuss several areas that need continued investigation.

Assessment of human locomotion by using an insole measurement system and artificial neural networks

A new method for measuring and characterizing free-living human locomotion is presented. A portable device was developed to objectively record and measure foot-ground contact information in every step for up to 24h. An artificial neural network (ANN) was developed to identify the type and intensity of locomotion. Forty subjects participated in the study. The subjects performed level walking, running, ascending and descending stairs at slow, normal and fast speeds determined by each subject, respectively. The device correctly identified walking, running, ascending and descending stairs (accuracy 98.78%, 98.33%, 97.33%, and 97.29% respectively) among different types of activities. It was also able to determine the speed of walking and running. The correlation between actual speed and estimated speed is 0.98, p< 0.0001. The average error of walking and running speed estimation is -0.050+/-0.747 km/h (mean +/- standard deviation). The study has shown the measurement of duration, frequency, type, and intensity of locomotion highly accurate using the new device and an ANN. It provides an alternative tool to the use of a gait lab to quantitatively study locomotion with high accuracy via a small, light and portable device, and to do so under free-living conditions for the clinical applications.

The Oswestry Disability Index

STUDY DESIGN

The Oswestry Disability Index (ODI) has become one of the principal condition-specific outcome measures used in the management of spinal disorders. This review is based on publications using the ODI identified from the authors' personal databases, the Science Citation Index, and hand searches of Spine and current textbooks of spinal disorders.

OBJECTIVES

To review the versions of this instrument, document methods by which it has been validated, collate data from scores found in normal and back pain populations, provide curves for power calculations in studies using the ODI, and maintain the ODI as a gold standard outcome measure.

SUMMARY OF BACKGROUND DATA

It has now been 20 years since its original publication. More than 200 citations exist in the Science Citation Index. The authors have a large correspondence file relating to the ODI, that is cited in most of the large textbooks related to spinal disorders.

METHODS

All the published versions of the questionnaire were identified. A systematic review of this literature was made. The various reports of validation were collated and related to a version.

RESULTS

Four versions of the ODI are available in English and nine in other languages. Some published versions contain misprints, and many omit the scoring system. At least 114 studies contain usable data. These data provide both validation and standards for other users and indicate the power of the instrument for detecting change in sample populations.

CONCLUSIONS

The ODI remains a valid and vigorous measure and has been a worthwhile outcome measure. The process of using the ODI is reviewed and should be the subject of further research. The receiver operating characteristics should be explored in a population with higher self-report disabilities. The behavior of the instrument is incompletely understood, particularly in sensitivity to real change.

Disability measurement in persons with back pain: a validity study of spinal range of motion and velocity

OBJECTIVE

To evaluate the criterion validity and responsiveness to change of spine kinematic variables to assess disability in patients with low back pain.

DESIGN

Blinded comparison between spine kinematic variables, Oswestry disability questionnaire scores, and work status.

SETTING

Multidisciplinary occupational rehabilitation clinic of a university hospital.

PATIENTS

Population-based cohort of 111 patients with subacute work-related back pain who were absent from regular work for more than 4 weeks because of back pain.

INTERVENTIONS

This study was part of a population-based randomized clinical trial. Patients were randomized to 4 different methods of management: usual care, rehabilitation, ergonomics, or rehabilitation and ergonomics.

MAIN OUTCOME MEASURES

Oswestry disability questionnaire, kinematic analysis of the spine during flexion and extension of the trunk, and work status were collected at weeks 4, 12, 24, and 52 after the back accident.

RESULTS

Kinematic variables were poorly to moderately related to work status and Oswestry questionnaire scores. Kinematic variables were also unresponsive to change in work status and Oswestry questionnaire scores over time.

CONCLUSION

Spine kinematics during flexion and extension of the trunk do not appear to be a valid measure of disability in patients with subacute and chronic back pain.

Discriminant analysis of human kinematic data: application to human lumbar spinal motion

A study was undertaken to determine the applicability of a multivariate discriminant technique in order to analyse human kinematic data, specifically lumbar motion during forward flexion. This method was used in an attempt to allow comparison of time-series data (three joint angles and three linear displacements) between groups of subjects. Results obtained from ten healthy subjects performing simulated abnormal styles of forward flexion indicate the feasibility and potential utility of this method in a clinical environment. Further investigations will be undertaken on clinical subjects to discriminate more effectively between healthy and pathological movements.