Image-based classification of bladder state using electrical impedance tomography.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Machine Learning

Background:

• Urinary incontinence (UI) affects millions, necessitating improved monitoring solutions.
• Current UI monitoring methods can be invasive, costly, or lack discretion.
• Electrical Impedance Tomography (EIT) offers a non-invasive approach to visualize internal body regions.

Purpose of the Study:

• To investigate the efficacy of machine learning classification for determining bladder states (full or not full) using EIT images.
• To develop a low-cost, discrete system for monitoring bladder fullness to aid UI management.
• To assess the potential of EIT-based machine learning for proactive UI alerts.

Main Methods:

• Utilized both numerical and experimental datasets of pelvic EIT images.
• Employed pixel-wise and feature extraction-based classification techniques.
• Applied principal component analysis, wavelets, and image segmentation for feature engineering.

Main Results:

• Achieved high classification accuracies, ranging from 77.50% to 100%.
• The optimal combination of principal component analysis and Gaussian radial basis function kernel support vector machine yielded 100% accuracy.
• This combination demonstrated the best balance between classification performance and computational resource usage.

Conclusions:

• Successfully performed the first machine learning classification of bladder EIT images.
• Demonstrated the potential of EIT combined with machine learning for effective bladder monitoring.
• This approach could significantly enhance the management of urinary incontinence.