As of right now, there are no reliable biomarkers for identifying Parkinson's Disease (PD) or monitoring its development. Here, using signals from nocturnal breathing, we created an Artificial Intelligence (AI) model to identify PD and follow its development. The model was assessed using data from numerous hospitals in the United States as well as numerous public datasets on a sizable dataset with 7,671 persons. On held-out and external test sets, the AI model can identify PD with an area under the curve of 0.90 and 0.85, respectively. The Movement Disorder Society Unified Parkinson's Disease Rating Scale, which is used to measure PD severity and progression, can also be used by the AI model. The AI model employs an attention layer that enables interpretation of its sleep and electroencephalogram predictions. Additionally, the model can detect breathing via radio waves that reflect off a person's body while they sleep to diagnose PD in the home environment touchless. Our study provides preliminary evidence that our AI model may be helpful for risk assessment before clinical diagnosis and shows the viability of objective, noninvasive, at-home evaluation of PD.
