PPMI integration
Baseline-first harmonization of 4,775 patients across 388 features covering MDS-UPDRS, UPSIT, REM sleep behavior disorder, biospecimens, and genetics.
Program outcomes & validation
We continually evaluate our pipelines across imaging, diffusion, wearables, and clinician-facing deployment. This page summarizes quantitative readouts, cohort coverage, and the evidence trail that underpins Progressive AI.
Cohort coverage
LRRK2 Consortium
2,958 participants inform genetic risk analyses and mechanism inference, highlighting the 1.89× elevated risk among G2019S carriers.
Wearable cohort
84 idiopathic Parkinson’s participants with synchronized CSFSAA labels, IMU-derived gait metrics, and comprehensive cognitive batteries.
Clinician pilots
ActionIntel and Free Motion prototypes evaluated with neurologists in weekly design reviews to pressure-test usability and interpretability.
Imaging classification performance
Automated SegFormer-based segmentation combined with DaT-SPECT features enables high-fidelity CSFSAA discrimination. Results below reflect cross-validated models trained on region-wise volumetrics and SBR-derived statistics.
| Method | F1-score | ROC-AUC |
|---|---|---|
| Nearest Neighbors | 0.960 | 0.900 |
| Random Forest | 0.950 | 0.858 |
| Neural Network | 0.950 | 0.843 |
| Gaussian Process | 0.944 | 0.837 |
| Naive Bayes | 0.935 | 0.907 |
Imaging alone yields AUC 0.70 ± 0.08; combining imaging with clinical and biologic features raises AUC to 0.93 ± 0.04.
Wearable-driven CSFSAA prediction
Histogram Gradient Boosting models trained on PPMI gait assessments quantify the contribution of each modality toward predicting CSFSAA status.
| Feature Group | Mean AUC ± SD |
|---|---|
| Gait & Arm Swing | 0.58 ± 0.16 |
| Demographics | 0.81 ± 0.14 |
| Clinical | 0.92 ± 0.07 |
| All Modalities | 0.93 ± 0.07 |
SHAP analysis surfaces HVLT retention, Epworth Sleepiness Scale, and right-arm jerk amplitude as leading contributors.
Mechanism-inference highlights
- Five motor phenotypes emerge from unsupervised clustering (Silhouette = 0.535), capturing differential striatal vulnerability.
- 50.2% hyposmia prevalence and 37.5% REM sleep behavior disorder prevalence confirm early cholinergic and brainstem involvement.
- Objective gait measures correlate with UPDRS-III severity (Spearman r = -0.301, p = 7.9×10⁻⁵, n = 166), grounding digital phenotyping in mechanism hypotheses.
Looking ahead
We are expanding evaluation to prospective, multi-site cohorts with emphasis on drift audits, fairness reporting, and regulatory documentation. Outcomes will be updated quarterly as we release new validation notes and open-source assets.