Figure 1
PRISMA Figure. This shows the outline of included studies as well as the reason for exclusion.
Table 1
Comprehensive table of included studies, number of patients, tremor types, data acquisition types, types of AI/ML, and diagnostic testing measures.
| STUDY NAME | TOTAL N | TREMOR TYPES AND N | DATA TYPES | AI METHODS | ACCURACY/SENSITIVITY/SPECIFICITY |
|---|---|---|---|---|---|
| Ai et al. 2007 | 26 | ET: 6 PD: 10 PT: 10 | Hand Accelerometer | BPNN | Accuracy: 92.9% |
| Ai et al. 2008 | 40 | ET: 10 PD: 15 PT: 15 | Hand Accelerometer | BPNN | Accuracy: 96.67% |
| Ai et al. 2011 | 25 | ET: 10 PD: 15 | Hand Accelerometer | SVM | Accuracy: 98% Sensitivity: 97.5% Specificity: 98.33% |
| Anandapadmanabhan et al. 2025 | 1077 | ET: 215 PD: 211 ET+: 208 DT: 365 CA: 78 | Archimedes’ Spirals (pen/paper) | CNN | Accuracy: 81% |
| Aracri et al. 2024 | 72 | ET: 32 PD: 40 | EMG | GB/xGB | Accuracy: 84.5% Sensitivity: 85% Specificity: 83.8% |
| Aubin et al. 2012 | 14 | ET: 7 PD: 7 | Inertial Measurement Unit | SVM | Accuracy: 85.7% |
| Balachandar et al. 2022 | 105 | ET: 49 PD: 30 DT: 26 | Hand Accelerometer (smartphone), Tremor Rating Scale | UMAP; LOOCV | Accuracy: 88% |
| Chandra Reddy et al. 2024 | 21 | ET: 9 PD: 5 CT: 7 | Video | CNN | Accuracy: 87.5% |
| Darnall et al. 2012 | 10 | ET: 3 PD: 6 ET/PD: 1 | Gyroscope, Archimedes’ Spirals (Digital), Tremor Rating Scale | DT, kNN, MP, NB, RF, SVM | Accuracy: 82% |
| Duque et al. 2022 | 39 | ET: 20 PD: 19 | Gyroscope (smartphone) | kNN, SVM | Accuracy: 77.8% Sensitivity: 75.7% Specificity: 80% |
| Engin et al. 2007 | 12 | ET: 3 PD: 9 | Hand Accelerometer | ANN | Accuracy: 91.02% Sensitivity: 92.97% Specificity: 87.13% |
| Ferreira et al. 2022 | 37 | ET: 10 PD: 27 | Hand Accelerometer, Gyroscope | DT, kNN, NB, RF, SVM | Accuracy: 100% |
| Ghassemi et al. 2016 | 24 | ET: 11 PD: 13 | Hand Accelerometer, EMG | SVM | Accuracy: 83% |
| Gonzalez et al. 2014 | 154 | ET: 34 PD: 120 | Hand position, velocity, acceleration | MP | Accuracy: 80% |
| Groznik et al. 2013 | 132 | ET: 52 PD: 46 Mixed: 34 | Archimedes’ Spirals (Digital) | ABML | Accuracy: 91% Sensitivity: 86% Specificity: 90% |
| Hossen et al. 2012 | 40 | ET: 20 PD: 20 | Hand Accelerometer, EMG | BPNN | Accuracy: 91.6% Sensitivity: 95% Specificity: 88.2% |
| Hossen et al. 2022 | 40 | ET: 20 PD: 20 | Hand Accelerometer, EMG | FF-BPNN | Accuracy: 92.5% Sensitivity: 98.47% Specificity: 100% |
| Ishii et al. 2020 | 50 | ET: 24 CA: 26 | Archimedes’ Spirals (Pen/Paper) | CNN | Accuracy: 70% Sensitivity: 44% Specificity: 79% |
| Jakubowski et al. 2002 | 174 | ET: 35 PD: 39 PT: 100 | Hand Accelerometer | MP | Accuracy: 97% |
| Kovalenko et al. 202 | 57 | ET: 13 PD: 42 Other: 2 | Video | GPC, GB/xGB, LR, RF, SVM | Accuracy: 77% |
| Lee et al. 2023 | 105 | ET: 18 PD: 87 | Video, Tremor Rating Scale | LSTM | Accuracy: 60% Sensitivity: 60% Specificity: 95% |
| Li et al. 2023 | 31 | ET: 12 PD: 19 | Hand Accelerometer | SVM; LOOCV | Accuracy: 100% Sensitivity: 100% Specificity: 100% |
| Lin et al. 2023 | 164 | ET: 80 PD: 84 | Inertial Measurement Unit (wearable), Task | LR; LOOCV | Accuracy: 84% Sensitivity: 85.9% Specificity: 82.1% |
| Locatelli et al. 2020 | 24 | ET: 7 PD: 17 | Hand Accelerometer | DT, DA, kNN, NB, SVM | Accuracy: 90.9% Sensitivity: 94.1% Specificity: 80% |
| Moon et al. 2020 | 567 | ET: 43 PD: 524 | Hand Accelerometer, Gyroscope (wearables) | DT, GB/xGB, kNN, LR, RF, SVM | Accuracy: 89% Sensitivity: 61% |
| Nanayakkara et al. 2025 | 66 | ET: 15 PD: 51 | Hand Accelerometer | LSTM | Accuracy: 95% |
| Nanda et al. 2015 | 2 | ET: 1 PD: 1 | Hand Accelerometer, EMG | FF-BPNN | NA |
| Oktay et al. 2020 | 40 | ET: 17 PD: 23 | Video | LSTM | Accuracy: 90% |
| Piepjohn et al. 2022 | 478 | ET: 305 PD: 173 | Hand Accelerometer, EMG | CNN, DT | Accuracy: 85.76% |
| Ranjan et al. 2020 | 27 | ET: 13 PD: 14 | Hand Accelerometer | kNN, RF, SVM; GMM, kM | Accuracy: 94.68% Specificity: 98.73% |
| Saad et al. 2024 | 43 | ET: 14 PD: 29 | Video | GB/xGB, SVM | Accuracy: 93% Specificity: 96% |
| Sanderson 2020 | 46 | ET: 12 PD: 34 | Inertial Measurement Unit | SVM | Accuracy: 92.4% |
| Seedat 2020 | 1039 | ET: 669 PD: 370 | Archimedes’ Spirals (Pen/Paper) | CNN | Accuracy: 92% |
| Shahtalebi 2020 | 162 | ET: 81 PD: 81 | Hand Accelerometer | PHTnet | NA |
| Shahtalebi 2021 | 162 | ET: 81 PD: 81 | Hand Accelerometer | DA | Accuracy: 95.5% |
| Skaramagkas 2020 | 15 | ET: 3 PD: 12 | Hand Accelerometer | DT, DA, EL, kNN, SVM | Accuracy: 100% |
| Skaramagkas 2021 | 15 | ET: 3 PD: 12 | Hand Accelerometer | DT, DA, EL, kNN, SVM | Accuracy: 100% |
| Spyers-Ashby 1999 | 57 | ET: 21 PD: 19 MS: 17 | Hand Accelerometer | kNN | Accuracy: 60% |
| Surangsrirat 2016 | 52 | ET: 20 PD: 32 | Gyroscope | SVM | Accuracy: 100% Sensitivity: 100% Specificity: 100% |
| Tang 2024 | 140 | ET: 53 PD: 87 | Hand Accelerometer, EMG | CAMN | Accuracy: 97.18% |
| Tavakkoli 2014 | 40 | ET: 20 PD: 20 | EMG | CNN, SVM | Accuracy: 95.75% |
| Teo 2024 | 143 | ET: 25 PD: 118 | Task | LSTM | Accuracy: 90% Sensitivity: 70% Specificity: 100% |
| Vescio 2023 | 40 | ET: 20 PD: 20 | Inertial Measurement Unit (wearables) | GB/xGB, RF | Accuracy: 92% Sensitivity: 96% Specificity: 87% |
| Weede 2024 | 339 | ET: 209 PD: 130 | Hand Accelerometer | CNN | Accuracy: 88.12% |
| Xing 2022 | 79 | ET: 28 PD: 51 | Hand Accelerometer, EMG | BPNN, CNN, GB/xGB, LR, RF, RRC, SVM | Accuracy: 85% Specificity: 64% |
| Yang 2020 | 26 | ET: 5 PD: 21 | Archimedes’ Spirals (Digital) | GRNN | Accuracy: 88.27% |
[i] Tremor Types: CA–Cerebellar Ataxia; CT–Cerebellar Tremor; DT–Dystonic Tremor; ET–Essential Tremor; ET+–Essential Tremor Plus; ET/PD–Essential Tremor/Parkinson’s Disease; Mixed–Mixed Tremor; MS–Multiple Sclerosis; Other–Other tremor, not otherwise classified; PD–Parkinson’s Disease; PT–Physiological Tremor.
Data Types: EMG – Electromyography; Inertial Measurement Unit – combines accelerometer, gyroscope, and magnetometer data in one device.
AI Types: Supervised Learning: ANN–Artificial Neural Network; BPNN–Back Propagation Neural Network; CAMN–Cross Attention Mechanism Network; CNN–Convolutional Neural Network; DT–Decision Tree; DA–Discriminate Analysis; EL–Ensemble Learning; FF-BPNN–Feed Forward Back Propagation Neural Network; GPC–Gaussian Process Classifier; GRNN–Generalized Regression Neural Network; GB/xGB–Gradient Boosting/eXtreme Gradient Boosting; kNN–k Nearest Neighbors; LR–Logistic Regression; LSTM–Long Short Term Memory; MP–Multilayer Perceptron; NB–Naive Bayes; RF–Random Forest; RRC–Ridge Regression for Classification; SVM–Support Vector Machine.
Unsupervised Learning: GMM–Gaussian Mixture Model; kM–k Means.
Semi-Supervised Learning: PHTnet–Patch-wise Hierarchical Transformer Network; UMAP–Uniform Manifold Approximation and Projection for Dimension Reduction.
Multiple: ABML–Argument-Based Machine Learning.
Validation: LOOCV–Leave One Out Cross Validation.
Figure 2
Conceptual overview of the current state and future directions for machine learning in tremor classification. Key future steps include conducting rigorous clinical trials, aggregating data into publicly available datasets, and developing an international, consortium-led smartphone application.