Radio Frequency-Enabled Cerebral Blood Flow Monitoring and Classification Using Data Augmentation and Machine Learning Techniques

Cerebral blood flow (CBF) signifies the rate at which blood circulates within the brain’s vascular network. CBF irregularities can lead to insufficient blood delivery to the brain, impacting cerebral metabolic processes. Consequently, this leads to gradual deterioration of neuronal health, potentially leading to cognitive decline, vascular dysfunction, dementia or stroke. Regular monitoring of CBF is critical for early detection of irregularities in the neurovascular system. The conventional neuroimaging techniques are costly, lack easy accessibility and necessitate significant supervision to operate. This paper proposes a novel Radio Frequency (RF) sensing system to effectively detect blood flow variations through backscattered RF signals. The non-invasive, cost-effective and portable sensing system features a novel miniaturized U-shaped antenna sensor integrated with eyeglasses for real-time monitoring. The sensor design is validated through microwave computational software and the fabricated sensing system is experimentally evaluated on an artificial brain model with an integrated arterial network of varying diameters. The results are measured on variable flow rates to accurately detect variations ranging from 10 to 90 mL/min. Machine Learning (ML) and Deep Learning (DL) methodologies are analyzed for classification into low, average and high CBF. A Gaussian noise feature data augmentation method is implemented on statistical feature and autonomous (AutoEncoders (AE) and Stacked AutoEncoders (SAE)) feature data. Ensemble bagged tree and Linear SVM offer a binary and multiclass classification accuracy of 96.2% and 88.9%, respectively on the statistical feature data augmented. Gaussian SVM has accuracies of 87.1% and 76.3% for binary and multi-class classification with SAE (32-16-32).