Table 1.

Summary of existing classification approaches for PD diagnosis.

Authors	Objectives	Sample Size	Features	Methods	Accuracy
Diego et al. (2018) [21]	Classify PD patients and HC subjects	388 subjects obtained from PPMI database	Morphological features extracted from DaTSCAN images with biomedical tests	SVM classifier with LOO-CV method	96%
Nicolas Nicastro et al. (2019) [24]	Distinguish PD patients from other parkinsonian syndromes and HC subjects	578 subjects (local database)	Semi-quantitative 123-FP-CIT SPECT uptake values	SVM with five-fold CV method	58.4%
Yang et al. (2020) [22]	Classify PD patients and HC subjects	101 subjects taken from PPMI dataset	Multimodel neuroimaging features composed of MRI and DTI with clinical evaluation	SVM, Random Forests, K-nearest Neighbors, Artificial Neural Network and Logistic Regression with ten-fold CV method	96.88%
Dotinga et al. (2021) [23]	Distinguish PD patients from non-PD subjects	210 subjects	SBR values computed from I-123 FP-CIT SPECT, age and gender	SVM with ten-fold CV method	95%
Lavanya Madhuri Bollipo et al. (2021) [25]	Classify early PD patients and HC subjects	600 subjects obtained from PPMI dataset	Clinical scores, SBRs values and demographic information	Incremental SVM with LOO-CV method	98.3%
Lavanya Madhuri Bollipo et al. (2021) [26]	Distinguish early PD patients from HC subjects	634 subjects taken from PPMI dataset	Motor, cognitive symptom scores and SBR values computed from DaTSCAN	SVR	96.73%
Diego Castillo-Barnes et al. (2021) [27]	Distinguish PD patients from HC subjects	386 samples selected from PPMI database	Morphological features computed from 123I-FP-CIT SPECT	SVM, Naive Bayesian and MLP with ten-fold CV method	97.04%