Table 1 Summaries of Studies Employing Machine Learning Algorithms for Movement Analysis and Back Pain Assessment.
From: Utilizing machine learning to analyze trunk movement patterns in women with postpartum low back pain
Authors | Year | Summary | Algorithm | Accuracy |
|---|---|---|---|---|
Nait Aicha et al.26 | 2018 | Machine learning, applied to accelerometer data gathered from a home environment, provides comparable accuracy to conventional models in identifying fall risks among older individuals. This method’s advantage is its independence from manually extracted features. Ultimately, this technique holds great potential in addressing societal challenges related to promoting active and healthy aging within the comfort of home | CNN, LSTM | (AUC = 0.75) |
Abdollahi, et al.27 | 2020 | This study tested a combination of a basic, inexpensive motion-capture sensor and the STarT questionnaire to make decisions in healthcare and telemedicine. The results showed that machine learning algorithms, primarily SVM, can differentiate between high-risk and low-risk NSLBP patients with above 75% accuracy. It was also found that tracking trunk movements could partially collect STarT questionnaire data implicitly. These results can help create an objective NSLBP evaluation tool using current technologies, which could have diagnostic and prognostic value. Creating a smartphone app for these tools and sharing quantifiable patient data with doctors, could revolutionize healthcare applications and significantly improve precision medicine | SVM and MLP | Accuracy levels of ~ 75% and 60% |
Rothstock, et al.28 | 2020 | The study’s key contribution is a clinically useful tool that can help physicians diagnose scoliosis and arrange treatments. The doctor receives assistance in making decisions and designing patient-specific brace treatments according to severity and therapeutic group categorization. The elements of the torso with high asymmetry distances (patches) can be exploited to design unique brace characteristics. Prospects include expanding the patient database and using more advanced neural networks (CNNs) in the context of deep learning | ANNs | 90% (SE: 80%, SP: 100%) for curve severity (mild vs. moderate-severe) and 50–72% for the ALS group |
Moniri, et al.29 | 2021 | Instead of the two lifting strategies used in current cutting-edge research, the best-unsupervised machine learning methodology based on Ward’s clustering correctly discriminated between four separate movement groups in persons with CLBP. The clustering result (four clusters) was confirmed using supervised machine learning using a Bayesian neural network with 97.9% accuracy. This promising method may assist in the more exact evaluation and rehabilitation of people with CLBP | CNN | 6.88% mean absolute percentage error and 3.72% standard deviation of absolute percentage error |
Phan, et al.30 | 2022 | Machine Learning Derived Lifting Techniques and Pain Self-Efficacy in People with Chronic Low Back Pain. Sensors | Bayesian neural network | 97.9% |
Thiry et. al.31 | 2022 | The goal of this study was to assess the usefulness of several machine learning algorithms and SampEn in identifying LBP situations. Using raw data from three IMUs and SampEn values collected during a B&R test, the findings demonstrated that the Gaussian NB ML algorithm performed better in discriminating CLBP patients from NLBP subjects than the SampEn discriminant values alone. This study found that supervised ML and a complexity evaluation of trunk variation in motion are beneficial in identifying CLBP situations, while simple kinematic markers are sensitive to the latter | Support Vector Machine, SVM | Achieving 79% accuracy |
Zahid Rao32 | 2024 | This study showed the creation of an active orthosis for people with impaired trunk control. The use of EMG and IMU data, along with a novel three-level categorization approach, produced encouraging results. Further study into feature selection and model optimization has the potential to improve results, promoting more freedom and well-being for wheelchair users. This study establishes a solid platform for upcoming assistive technologies in trunk rehabilitation | ES and two with KNN | 95.44–87.0% |
Doaa A. Abdel Hady and Tarek Abd El-Hafee (The Proposed Work) | 2024 | In this research, machine learning was utilized to analyze and predict low back pain in postnatal women based on trunk movement. A dataset from 100 postpartum women was used, including those with and without low back pain. The optimal regression model was the Optimized optuna Regressor, while the Basic CNN and Random Forest Classifier achieved almost perfect results in pain classification. Key indicators of pain were identified as pain level, range of motion, and average movements. Though the study’s dataset was limited, the machine learning models provided insightful and accurate analysis of the factors contributing to low back pain. The findings underscore the potential of machine learning in improving low back pain risk assessment and tailoring treatment strategies | Optuna, CNN, and Random Forest Classifier | (MSE) of 0.000273, (MAE) of 0.0039, and an R2 score of 0.9968. In classification, the Basic CNN and Random Forest Classifier both attained near-perfect accuracy of 1.0 and F1-score of 1.0, |
