Table 13 A comparison of movement data papers
Article | Main idea | Tools | Applied algorithms | Advantages | Disadvantages |
|---|---|---|---|---|---|
PD diagnosis and severity evaluation using MCSVM | • MATLAB | • MCSVM • SVM kernel functions (linear, polynomial, cubic, and quadratic) | • High accuracy • High specificity • High sensitivity | • Lack of non-motor symptom evaluation • Not analyzing metrics such as recall and cost | |
DL and neuro-fuzzy model for PD detection | • MATLAB | • DBN • KNN • ANFIS • EM • PCA | • Low time complexity • High accuracy | • Not evaluating parameters such as sensitivity and recall | |
FOG prediction in PD using ResNeXt | • Python (Pytorch) | • ResNeXt • SMOTE • Adam optimizes | • High accuracy • High sensitivity • high specificity | • Low scalability | |
Analyze real-world gait tests in PD patients | • Not mentioned | • SDTW | • High F1-score • High recall • High precision | • Lack of analysis of gait trials | |
Analysis of sEMG signals and hybrid DTL for diagnosing PD | • MATLAB | • CNN • SVM • SGD • Propagation | • High accuracy • High specificity • High sensitivity | • Low scalability | |
Assessing PD severity via the EnKNN approach | • Python | • EnKNN | • High accuracy | • Low scalability | |
Real-time FOG detection in PD using CNN | • MATLAB • Python • Keras-flops | • CNN | • Low computational complexity • Low processing time • High performance • Reducing memory usage • High AUC • High predicting ability | • No integration with a standalone device for home environment • Utilization of raw input data | |
Early PD detection using wearable sensors and ML | • Not mentioned | • LightGBM • RF | • High precision • High F1-micro • High AUC | • Lack of additional data types such as video/images | |
CNN for PD and AD classification | • R programming • Python | • Multi-layer CNN • LDA • MLP | • High accuracy | • Lack of evaluation on imbalanced datasets • Low scalability | |
PD symptom detection through video analysis | • Python (Scikit-learn) • OpenCV | • LR • XGBoost • RF • SVM • Gaussian process classifier | • High F1-score | • Lack of consideration of datasets with varied disease • Not examining the data with other models, such as CNN | |
Handwriting analysis via CNN for PD diagnosis | • Not mentioned | • CNN | • High accuracy | • Not investigating alternative architectures • Low scalability | |
Movement management in PD patients using DL | • Python | • DNN | • High accuracy | • Lack of minimizing the model effect | |
DL method to distinguish MS from PD via gait | • Python (PyTorch) | • CNN • RNN • MS-ResNet | • High accuracy • High AUC | • Low scalability | |
Neural network–based early PD detection via gait data | • Not mentioned | • Neural network | • High accuracy | • Training model with a limited number of patients • Low scalability | |
Motor symptom–based PD detection using ML | • MATLAB • Python | • Lasso • LR • RF • DT • SVM • KNN • XGBoost • Linear Regression | • High accuracy • High AUC (ROC) | • Low scalability | |
PD severity assessment via DL on movement data | • Python | • CNN-BGRU | • High accuracy | • Not analyzing parameters such as AUC (ROC) and specificity • Lack of clinical validation | |
ML-based detection of PD using upper limb motion | • MATLAB | • DT • RF • KNN • SVM • NB | • High accuracy • High sensitivity • High specificity • High AUC (ROC) | • Low scalability | |
Analyzing copied figures with CNN to detect PD | • Python (NumPy, Pandas) | • CNN | • High accuracy • High specificity | • Low scalability | |
Analyzing VGRF gait data via ML to detect PD | • Not mentioned | • SVM • KNN • NB • DT • ELA | • High accuracy | • Low scalability • Neglecting all gait signals but VGRF | |
Learning architecture for PD diagnosis | • Python • CUDA • cuDNN | • CNN • ARR • XGBoost • SMOTE | • High accuracy • Low training time | • Low scalability | |
Gait-based PD detection and stages with ML models | • Python (NumPy, Matplotlib, Scikit-learn, Pandas, Seaborn) | • NB • SVM • DT • MLP • LR • RF • SMOTE | • High accuracy • High precision • High AUC (ROC) | • Low scalability • Not evaluating other motor and non-motor symptoms • Lack of tremor analysis in gait classification | |
Extracting diagnostic features from spiral drawings using ML | • Python (Scikit-learn) | • LR • SVM • KNN • DT • RF • AdaBoost • SVM-RFE | • High predicting ability • High specificity • High accuracy • High sensitivity | • Low scalability • Lack of symptom severity assessment • Not evaluating other tasks related to handwriting and drawing | |
Unsupervised uTUG-based gait assessment for PD using ML | • GroupKFold • GridSearchCV • Python (Scikit-learn) | • NB • SVM • RF | • High accuracy • High recall • High sensitivity • High F1-score • Not requiring manual annotation | • Not evaluating adverse drug reactions • Lack of additional sensor data • Incomplete evaluation of at-home completion time | |
Balanced ensemble learning for PD diagnosis utilizing KD dataset | • R programming | • XGBoost • KNN • NB • LSTM • MLP • SVM | • High sensitivity • High specificity • Ease of integration with conventional desktops • High robustness • High AUC | • Not investigating diseases that affect typing quality • Not assessing the impact of factors such as age, emotional tension, and keyboard layout experience on typing • No evaluation of wearable and mobile sensors for improved data collection • Not evaluating the severe level of the disease | |
Ensemble DT and gait features for PD detection | • Not mentioned | • RF • GB • DT | • High accuracy • High F1-score • High sensitivity • High specificity • High precision | • Low scalability | |
ML-based PD diagnosis using gait and movement data from wearable sensors | • MATLAB | • Random under-sampling boosting • Neighborhood component analysis • mRMR • RF • DT | • High sensitivity • High specificity • High AUC | • Low scalability • Lack of generalization assessment to other motor disorders • Lack of evaluation of motor fluctuations | |
Kinematic handwriting features and ML for PD diagnosis | • Python | • RNN • LSTM • BLSTM • Adaboos • BRF • SVM • LDA • PCA • Bayesian optimization algorithm • Adam optimizes | • High accuracy • High precision • High recall | • Not analyzing diverse handwriting datasets • Low scalability | |
PD detection by using handwriting and neural network | • Python (Tensorflow) | • NB • RF • DT • LR • KNN • GBDT • CNN • BLSTM • LSTM | • High accuracy | • Not expanding image datasets sufficiently | |
ML-based PD diagnosis by analyzing exercise effectiveness | • Python | • PCA • ICA • MDS • RF • LR • NB • Boosted trees • KNN • Stacked ensemble model | • High AUC (ROC) • Low hospitalization cost • Reduce diagnosis time | • Small and imbalanced dataset • Sensors with limited battery life • Wireless connection problems • Lack of comfort for people with advanced PD to wear sensors | |
PD detection via CNN based on daily gait patterns | • Python | • CNN | • High accuracy • High AUC | • Not distinguishing the level of PD • Not evaluating other forms of movement except walking | |
Monitoring PD motor symptoms using ML methods | • Python • MobilityLab software | • LR • RF • PCA | • High RMSE | • Low scalability | |
PD detection via ML approaches | • Python | • SVM • FNN • CatBoost • BOSS • XceptionTime | • High accuracy | • Small sample size for validation • One-Time clinical assessment | |
PD diagnosis using keystroke dynamics data | • Not mentioned | • MFDFA • CNN | • High accuracy • High sensitivity • High specificity | • Low scalability | |
LSTM-based classification of PD using walking data | • Not mentioned | • LSTM • MCOA | • High accuracy | • Low scalability |
