Supervised Learning Classifiers for Electrical Impedance-based Bladder State Detection.

Area of Science:

• Biomedical Engineering
• Machine Learning Applications
• Medical Device Technology

Background:

• Urinary Incontinence (UI) affects over 200 million globally, significantly reducing quality of life.
• Current management strategies for UI can be invasive or inconvenient.
• Technological solutions for real-time bladder state detection are needed to provide timely alerts and improve patient care.

Purpose of the Study:

• To investigate the feasibility of using supervised machine learning (ML) classifiers for bladder state detection ('full' or 'not full').
• To analyze the effectiveness of electrical impedance measurements for non-invasive bladder monitoring.
• To compare the performance of Support Vector Machines (SVM) and k-Nearest Neighbors (k-NN) classifiers in this application.

Main Methods:

• Electrical impedance data was collected from computational models and a realistic pelvic phantom.
• Multiple datasets were created with varying noise levels to simulate real-world conditions.
• 10-Fold cross-validation was employed to classify bladder states using SVM and k-NN algorithms.

Main Results:

• Classification accuracies ranged from 73.16% to 100% across different datasets.
• Both SVM and k-NN demonstrated effectiveness in classifying bladder states.
• Higher noise levels and bladder volumes near the 'full'/'not full' threshold were identified as key factors influencing misclassification.

Conclusions:

• Machine learning applied to electrical impedance measurements shows significant promise for bladder state detection.
• This approach could lead to novel, non-invasive devices to assist individuals with urinary incontinence.
• Further research is warranted to optimize the technology for clinical application and widespread use.