Researchers are investigating whether rapid, smartphone-based motor tests could serve as a feasible alternative to dopamine transporter (DaT) SPECT scans, a type of imaging procedure typically performed to identify dopaminergic impairments in Parkinson’s disease (PD). The study included participants with PD, isolated REM sleep behavior disorder (iRBD), and healthy controls. It assessed whether digital motor signals recorded with smartphone sensors reflect anomalies that are generally visible on DaT imaging. Although DaT scans remain the gold standard, they are costly, require specialized equipment, and expose patients to ionizing radiation.
The researchers aimed to evaluate the ability of these smartphone-based motor features, alone or in combination with the established Movement Disorder Society-Unified Parkinson’s Disease Rating Scale Part III (MDS-UPDRS-III), to detect abnormal DaT scans and provide a more accessible, scalable screening tool.
The dataset consisted of 100 DaT scans from 93 subjects (5 HCs, 49 iRBD, 39 PD), all with smartphone measurements collected within one year of the imaging. Among them, 52 scans were normal, and 48 were abnormal. MDS-UPDRS-III scores were significantly different between the two groups (p <0.00001), and females had a higher prevalence of abnormal scans (p = 0.00001).
Smartphone features characterizing gait, tremor, voice, dexterity, and other motor functions (a total of 1057 features) were used to train machine learning models. The highest-performing XGBoost classifier, using only smartphone features, achieved an AUC of 0.80, comparable to that of MDS-UPDRS-III alone (AUC 0.81). Combining the MDS-UPDRS-III with the most popular 500 smartphone features further improved performance, yielding an AUC of 0.84 (95% CI: 0.75-0.92). The same was observed in logistic regression: the MDS-UPDRS-III alone had an AUC of 0.83, and the combined model reached 0.85. Sex-adjusted models produced AUCs of 0.82 for XGBoost and 0.85 for logistic regression. Feature importance analyses consistently identified gait, rest tremor, and voice as the most influential contributors.
In a pre-screening sensitivity analysis excluding participants with moderate-to-severe PD (MDS-UPDRS-III ≥ 33), model performance decreased across all approaches. MDS-UPDRS-III was most effective in this less severe cohort (AUC 0.83). Still, the combined model showed lower variance across folds, suggesting more consistent performance at the cost of slightly reduced discriminative ability.
To explore whether smartphone data could provide a quantitative measure of dopaminergic loss, regression models were developed to predict specific binding ratios (SBRs) in four striatal regions: left and right caudate and left and right putamen. Smartphone-only XGBoost regressors achieved modest accuracy, while MDS-UPDRS-III alone performed better. Although the combination of smartphone and clinical data always reduced prediction error. The best performance was observed in the right putamen, where the combined model yielded RMSE = 0.49 and R² = 0.56, indicating a moderate correlation with the actual SBR values. Models based solely on smartphone features tended to underestimate higher SBR values, likely due to data imbalance.
Correlation analysis supported patterns observed in neurobiological studies. MDS-UPDRS-III scores showed Spearman’s correlations of -0.64 with right putamen SBR and -0.50 with the top smartphone tremor feature (Entropy Z-axis). Previous studies have demonstrated that bradykinesia, gait and posture impairment, and other midline symptoms are strongly associated with dopaminergic depletion; smartphone-based motor indices showed comparable associations.
Overall, these results indicate that smartphone-based motor testing can correlate with in-clinic assessments and has significant predictive value for dopaminergic impairment. Although not a replacement for DaT SPECT, smartphone testing, particularly when combined with clinical evaluation, may serve as a scalable pre-screening instrument to identify the disease at earlier stages, support better triage, and support recruitment for disease-modifying trials. Clinical adoption will require further validation in larger and more heterogeneous cohorts.
References: Gunter KM, Groenewald K, Aubourg T, et al. Smartphone-based prediction of dopaminergic deficit in prodromal and manifest Parkinson’s disease. npj Digit. Med. 2025. doi:10.1038/s41746-025-02148-2
