Textural feature based intelligent approach for neurological abnormality detection from brain signal data

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Diagnosing neurological diseases from electroencephalography (EEG) data is challenging due to large, complex time-series data.
• Current EEG classification methods are often disease-specific, lacking cost-effectiveness and a unified diagnostic framework.
• Visual analysis of EEG by neurologists is time-consuming, subjective, and prone to errors.

Purpose of the Study:

• To develop a unified machine learning (ML) based data mining technique for categorizing multiple neurological abnormalities from EEG signals.
• To create a cost-effective and efficient system for identifying diverse neurological disorders using a single EEG classification framework.

Main Methods:

• Pre-processing of EEG signals, including noise/artifact removal and normalization.
• Generation of time-frequency spectrogram images from segmented EEG data using Short-Time Fourier Transform.
• Extraction of histogram-based textural features, followed by Principal Component Analysis for feature selection.
• Classification of selected features into different disease classes using four distinct ML classifiers.

Main Results:

• The developed ML-based method demonstrated potential in classifying various types of neurological abnormalities from EEG data.
• The system was successfully tested on four real-time EEG datasets, validating its performance.
• The approach shows promise for a unified system to identify diverse neurological abnormalities.

Conclusions:

• The proposed machine learning technique offers a viable and unified approach for the multi-class classification of neurological diseases using EEG.
• This method can significantly aid in the objective and efficient diagnosis of various brain signal abnormalities.
• The study highlights the potential of advanced data mining and ML techniques in modern neurological diagnostics.