Cerebral Blood Flow Classification Dataset Using Ultra-Wideband RF Sensors with Statistical and Autoencoder-Based Features

Experimental Setup:

Cerebral blood flow data was collected using Radio Frequency (RF) sensors operating in the Ultra-Wideband (UWB) frequency range to enable non-invasive monitoring of cerebral hemodynamics. The experimental setup consisted of an RF transmitter (Tx) and receiver (Rx) positioned at a fixed distance to optimize signal transmission and minimize interference. These sensors captured variations in blood flow by detecting subtle physiological changes as RF signals penetrated the skull and biological tissues.

Feature Extraction and Augmentation:

The recorded RF signals were stored in .CSV format and processed for feature extraction. Statistical features were manually extracted to capture variations in cerebral blood flow, while autonomous feature extraction was performed using a Stacked Autoencoder (SAE) with a structure of 32-16-32. To enhance the feature dataset size, Gaussian noise-based augmentation with a mean of 0.2 and variance of 0 was applied to both statistical and autonomously extracted features. This augmentation expanded the feature space, improving model generalization.

Classification and Data Structure:

The dataset supports both binary and multi-class classification.

• Binary Classification: Blood flow values of 10 ml and 90 ml are labeled as abnormal, while 50 ml is labeled as normal.
• Multi-Class Classification: The dataset consists of six classes, distinguishing between Arterial Network A and Arterial Network B:
  • Arterial Network A: 10ml-A, 90ml-A, 50ml-A
  • Arterial Network B: 10ml-B, 90ml-B, 50ml-B

The dataset is organized into two main folders:

1. Statistical Feature Data:
   • Contains two .xlsx files:
     • data_0.2GaussianNoise-BinaryClass.xlsx
     • data_0.2GaussianNoise-Multiclass.xlsx
   • The statistical features include mean, standard deviation, quartile deviation, range, skewness, and kurtosis.
2. Stacked Autoencoder (SAE) Data:
   • Contains three .xlsx files:
     • Small or Arterial Network A – Binary Class: Represents binary autonomous feature data extracted using the SAE (32-16-32) and expanded via Gaussian noise augmentation.
     • Big or Arterial Network B – Binary Class: Contains feature data acquired using the same methodology.
     • SAE  Multiclass Feature Data Gaussian Augmented: Represents six-class autonomous feature data expanded through Gaussian noise augmentation.

Use:

This dataset serves as a foundation for developing machine learning models for cerebral blood flow classification, offering potential applications in non-invasive cerebrovascular health monitoring and the early detection of neurological disorders.