Table 1.
Related studies summarization concerning PD
| Reference | Year | Approach | Dataset | Dataset type | Pros. | Cons. | Best accuracy |
|---|---|---|---|---|---|---|---|
| Pereira et al. Pereira et al. 2015) | 2015 | ML using NB, SVM, and OPF | HandPD | Image | Proposing “HandPD” dataset | Achieved accuracy is low | 78.9% using NB |
| Pereira et al. Pereira et al. 2016b) | 2016 | CNN | Proposing an extension to the “HandPD” dataset using signals from a smartpen from meander and spiral drawings | (1) The use of an imbalanced dataset with more healthy samples and (2) the usage of tablet-based devices requires specific conditions for good quality | 80.19% | ||
| Pereira et al. Pereira et al. 2016a) | 2016 | Metaheuristics + CNN | Usage of metaheuristic algorithms to tune the hyperparameters | The usage of imbalanced dataset with more healthy samples | 90.39% | ||
| Pereira et al. Pereira et al. 2018) | 2018 | CNN | (1) CNN is applied for learning features from handwritten dynamics and (2) proposing “NewHandPD” dataset extracted by the use of a smartpen | Process of the time-series data in a black-box manner | 95% | ||
| Senatore et al. Senatore et al. 2019) | 2019 | CGP | The usage of Cartesian Genetic Programming to provide explicit classification rules | Poor results for spiral images | 72.36% | ||
| Impedovo et al. Impedovo 2019) | 2019 | SVM with a linear kernel | PaHaW | Usage of velocity signals | Useful in online handwriting only | 98.44% | |
| Naseer et al. Naseer et al. 2020) | 2020 | CNN using AlexNet | (1) The usage of fine-tuned pretrained models and (2) the usage of k-fold cross-validation | (1) No consideration of dimensionality reduction and (2) vulnerability to acoustic conditions | 98.28% | ||
| Kamran et al. Kamran et al. 2021) | 2021 | CNN using AlexNet, GoogLeNet, VGG, and ResNet | HandPD, NewHandPD, and Parkinson’s Drawing datasets | (1) The usage of several datasets and (2) high achieved accuracy. | Poor accuracy in case of scratch CNN | 99.22% using AlexNet | |
| Sakar et al. Sakar et al. 2013) | 2013 | SVM, KNN | Speech data | Voice | Proposal of voice dataset for Parkinson’s disease | Results are biased | 77.5% |
| Caliskan et al. Caliskan et al. 2017) | 2017 | DNN | OPD and PSD | Remote diagnosis ability | Low accuracy | 93.79% | |
| Tuncer et al. Tuncer and Dogan 2019) | 2019 | SVM, 1NN, DT, and logistic regression | Vowel | Gender classification is taken into account | The usage of small data | 97.62% by 1NN | |
| Zahid et al. Zahid et al. 2020) | 2020 | AlexNet | pc-Gita | (1) The usage of deep features of speech and (2) proving that pronunciation of vowels are sufficient in diagnosis | Poor accuracy for isolated words | 99.7% |