TABLE 4.
The pathological gait.
| Reference | Algorithm/Best Accuracy | Data Collection/Input | Pathology/Task Output |
|---|---|---|---|
| Lu et al. (2020) | SVM with PCA 88.89% | Kinect camera with image rectification | Automatic depression detection |
| Khodabandehloo et al. (2021) | HealthXAI CART | Partial CASAS dataset, 192 subjects: 19 PwD, 54 MCI | Numerical score and explanation of the decline of cognitive functions of the elderly |
| Iosa et al. (2021) | ANN 93.9% | IMU at the waist belt, N = 33, HC = 17, stroke = 16 | Stroke prognostic tool, able/unable to return to work |
| Flagg et al. (2021) | Bidirectional GRU | GaitNDD, GaitPDB. Streaming of live and historical GRFs | ND: PD, HD, ALS gait normality analysis |
| Costilla-Reyes et al. (2021) | Novel DNN with t-SNE F-score: 97.33% | Own dataset: UOM-GAIT-69. Tomography floor sensor raw data. N = 69, healthy normal/fast/dual-task | Age-related differences in healthy adults undertaking dual tasks |
| Zhu et al. (2020) | RF with IAFSA RMSE = 0.073 | 3 patients with knee replacement. Public dataset/challenge2 | Knee joint impairment KFC prediction |
| Zhou et al. (2020) | Kernel PCA with SVM, RF, ANN: 90% | N = 239, young = 57, old healthy = 55, 127 = old-geriatric condition | Geriatric condition |
| Zhang et al. (2020a) | Deep CNN AUC = 0.87 | DREAM PDDB Challenge | PD vs healthy gait; Large scale screening |
| Zeng et al. (2020) | RBF neural network with DL 95.61% | Kinematic modeling using a biped, N = 43 participants. Mocap, and force plates to test the model | Chronic unilateral ACL deficiency. Classify ACL-D/ACL-I knees |
| Pepa et al. (2020) | Novel FL Sp = 95.2%, Se = 84.9% | Smartphone data | Real time, interpretable FoG detection |
| Lasselin et al. (2020) | MLR | N = 19, lipopolysaccharide-induced inflammation. Kinect camera data | Effects of inflammation on human gait |
| Kaur et al. (2020) | LR, SVM, RF | 4 people with MS. GRFs from instrumented treadmill | GML4MS framework, HC/MS mild and moderate classifier |
| Bhattacharya et al. (2020) | ST-GCN and CVAE 88% | 4,277 human gaits in video and synthetic gaits by novel STEP-Gen | Emotion classification: happy, sad, angry, or neutral |
| Li et al. (2019) | WeedGait, by LSTM and SVM 92.1% | N = 10, smartphone data | assesses marijuana-induced gait impairment passively, warns against DUIM online |
| Guo et al. (2019) | SVM and BiLSTM | N = 16, light-weight telepresence robot equipped with a single RGB-D camera with no additional sensing feedback | normal, in-toeing, out-toeing, and drop-foot gait |
| (Zhang et al., 2019b) | ANN (a = 50) 93.5% | N = 200, 8-camera mocap and 3 force platforms | Gait classification for CP patients with spastic diplegia |
| Sato et al. (2019) | ST-ACF DTW, KNN with OpenPose | CASIA-B dataset. Frontal videos of two PD patients | Quantifying normal and Parkinsonian gait features from home movies |
| Fang et al. (2019) | RF 91.58% | 95 graduate students. 52 score-depressed, 43 HC. Two MS Kinect cameras | Depression analysis |
| Acosta-Escalante et al. (2018) | Logitboost & RF 94.5% on raw data | N = 14, HD = 7, HC = 7. Smart phones (iPhone 5S) affixed to both ankles | HD gait classification |
| Ye et al. (2018) | ANFIS/PSO with LOOCV. 94.44% | 64 subjects, ALS = 13, PD = 15, HD = 20, HC = 16, ND Public dataset. Force-sensitive switches are placed on subjects’ shoes. | Classification of Gait Patterns in Patients with various ND |
| Pulido-Valdeolivas et al. (2018) | RF with DTW | 26 HSP and 33 healthy children. Optokinetic IGA system | Monitoring HSP progression and personalizing therapies |
| Wan et al. (2018b) | DMLP. 97.9% | N = 50, phone worn on the waist. Biomedical voice recordings (UCI dataset) and smartphone 3-axial acceleration | Analyze speech and movement data captured by smartphone to estimate the severity of PD |
| Hasan et al. (2018) | ANN, SVM with SWDA 93.3% | 3D GRF data of 60 children: 30 ASD and 30 typically developing | Identifying ASD Gait |
| Cui et al. (2018) | SVM w/PCA 98.21% | N = 42, 21 post-stroke, 21 HC MT, GRF and EMG | Recognition and Assessment of PSH Gait |
| Ajay et al. (2018) | DT. 93.75% | 49 YouTube videos of varying resolution. Video obtained through any pervasive devices | PD gait classification |
| Arifoglu and Bouchachia, (2017) | LSTM HAR: 96.7% AAD: 91.43% | Public dataset collected in 3 households through environmental sensors (Van Kasteren et al., 2011) | HAR and AAD for elderly people with dementia |
| Bilgin, (2017) | LDA, NBC. 90.93% | GaitNDD. Force-sensitive resistors. 3 ALS, 15 PD, 20 HD, and 16 HC | Classification of ALS among other ND diseases and healthy subjects |
| Dolatabadi et al. (2017) | GPLVM-thold and KNN-DTW F1-score > 0.94 | N = 40, HC = 20, mobility impared = 20. Two Kinect sensors | Discriminate between healthy and pathological gait patterns because of stroke or ABI |
| Shetty and Rao, (2016) | SVM with Gaussian RBF 83.3% | GaitNDD. GRF measurements. N = 64, 15 PD, 18 HD, 13 ALS, 16 HC | Distinguish PD gait from HD, ALS, and HC |
| Procházka et al. (2015) | NBC 94.1% | N = 51, 18 PD, 18 HC - age-matched, and 15 young HC. MS Kinect Image and depth | PD diagnosis |
| Wahid et al. (2015) | RF with MR normalization. 92.6% | N = 49: PD = 23 HC = 26. 15 Reflected markers, 2 force platforms | PD diagnosis and management using normalized spatial-temporal gait |
Legend: Decision Tree (DT), K-Nearest Neighbors (KNN), Center for Advanced Studies in Adaptive Systems (CASAS) (Cook et al., 2015), Classification and Regression Trees (CART), Gated Recurrent Unit (GRU), Root Mean Square Error (RMSE), Receiver-Operating Characteristic (ROC), ACL Deficient (ACL-D), ACL-intact (ACL-I), Radial Basis Function (RBF), Deterministic Learning (DL), Multivariable Linear Regression (MLR), Multiple Sclerosis (MS), Gait data-based ML framework for MS prediction (GML4MS), Linear Regression (LR), Abnormal Activity Detection (AAD), Gaussian Process (GP) Latent Variable Models (GPLVM), OpenPose (Cao et al., 2017).
Datasets: CASIA-B (Yu et al., 2006), Gait in Neurodegenerative Disease Database (GaitNDD) (Hausdorff et al., 2000), Gait in Parkinson’s Disease (GaitPDB) (Goldberger et al., 2000), CASAS (Cook et al., 2015), 2 https://simtk.org/projects/kneeloads, ND Public Dataset (Hausdorff et al., 2000).