PD (Parkinson’s Disease) is the fastest-growing neurological illness worldwide, hence early diagnostic biomarkers are needed. No medications reverse or stop PD progression. PD is diagnosed based on motor abnormalities such tremors and rigidity.
The Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRaS) is used to measure PD progression qualitatively.
Some PD biomarkers, including cerebrospinal fluid, blood biochemistry, and neuroimaging, show promise in early diagnosis. These costly techniques need multispecialty hospitals, rendering them unsuitable for early diagnosis or PD progression tracking.
James Parkinson originally linked PD to respiration in 1817. PD patients have degraded brainstem areas that govern breathing, making breathing a suitable risk assessment feature for clinical diagnosis. Respiratory symptoms appear before PD motor symptoms.
In this study, researchers assessed an AI-based study model on 7,671 people utilising public datasets and U.S. hospitals.
The model collected data in two ways. First, a breathing belt was worn overnight to track breathing signals. This approach produced the breathing belt dataset.
The second contactless approach collected respiration data using a radio sensor that emitted a low-power radio signal in the individual’s bedroom. This was wireless data.
The model trained to predict nocturnal breathing-based qEEGs. Subjects who trained the neural network were not tested.
For PD identification, the team used k-fold cross-validation (k=4) and leave-one-out validation. Training and testing the model on medical centre data determined cross-institution prediction.
AI model detected PD from one night of nocturnal breathing (ROC curve). The AI model recognised PD with excellent accuracy, but researchers wanted to see if merging data from numerous nights in the same test subject enhanced its accuracy. All nights’ model prediction scores were computed using wireless data.
Test-retest reliability was measured by averaging predictions over consecutive nights. The AI model’s capacity to create a PD severity score correlated with the MDS-UPDRS.
Patients with PD averaged 69.1 years old, and 27% were women. The control group had 6,914 participants, 30% of whom were women aged 66.2.
Each person’s longitudinal data spanned up to a year. 11,964 nights and nearly 120,000 hours of nocturnal breathing signals from 757 PD patients were studied.
Breathing belt datasets lacked MDS-UPDRS and H&Y scores. Wi-Fi datasets have MDS-UPDRS and H&Y scores.
The study model has an AUC of 0.889, a sensitivity of 80.22%, and a specificity of 78.62%, all with a 95% confidence interval (CI). For nights evaluated using a wireless dataset, the model reached an AUC of 0.906 with 86.2% sensitivity and 82.8% specificity.
PD prediction score ranges from 0 to 1. When the AI model’s PD score surpasses 0.5, a person has PD. Researchers utilised each subject’s median PD score to diagnose them.
Combining nights for each individual boosted PD diagnosis sensitivity and specificity to 100% for PD and control subjects. Using multiple nights of the same patient raised dependability to 0.95 in 12 nights.
AI models’ severity predictions correlated strongly with MDS-UPDRS, demonstrating that the model identified PD disease severity accurately. Since the model predicted three MDS-UPDRS subparts with R values of 0.84, 0.91, and 0.93, it captured non-motor and motor PD symptoms.
The AI-based method reported in this paper is a promising PD diagnostic and progression biomarker. The model was objective, non-obtrusive, affordable, and could measure nighttime breathing at home.
About 40% of PD sufferers don’t see a specialist because they’re concentrated in urban locations. For such instances and those at high risk of PD, including those with leucine-rich repeat kinase 2 gene mutation, an AI system could be used for passive tracking at home. This technology may also provide regular input to the patient’s doctor, who might validate the results via telehealth or in-person visit.
AI breakthroughs can help medicine by addressing unresolved neuroscience difficulties and giving new clinical insights for diagnosing and tracking PD progression.
- Yang, Y., Yuan, Y., Zhang, G., et al. (2022). Artificial intelligence-enabled detection and assessment of Parkinson’s disease using nocturnal breathing signals. Nature Medicine. doi:10.1038/s41591-022-01932-x