Brain haemorrhage detection using a SVM classifier with electrical impedance tomography measurement frames.

Area of Science:

• Medical Imaging
• Machine Learning
• Biomedical Engineering

Background:

• Brain haemorrhages necessitate rapid and precise diagnosis for effective treatment.
• Electrical Impedance Tomography (EIT) offers a potential non-invasive method for detecting intracranial abnormalities.

Purpose of the Study:

• To evaluate the efficacy of Support Vector Machine (SVM) classifiers in identifying brain haemorrhages using EIT data.
• To assess the influence of various factors on classifier performance in both numerical and physical models.

Main Methods:

• Development of a 2-layer head model with simulated haemorrhages for numerical simulations and physical phantoms.
• Training and testing linear SVM classifiers using EIT measurement frames from head surface electrodes.
• Investigating the impact of noise, lesion characteristics, electrode placement, and anatomical variations.

Main Results:

• Linear SVM achieved high accuracy (90%+) in numerical models, detecting lesions as small as 5 ml, independent of location.
• Phantom models showed lower maximal sensitivity and specificity (75%) compared to numerical simulations.
• More complex classifiers like RBF SVM and neural networks demonstrated improved detection accuracy.

Conclusions:

• SVM classifiers applied to EIT data represent a novel approach for brain haemorrhage detection.
• Translating numerical model success to clinical settings requires addressing anatomical variations and optimizing classifiers.
• Further research is needed to develop robust EIT-based diagnostic tools for brain haemorrhages.