A Pilot Study: Detrusor Overactivity Diagnosis Method Based on Deep Learning.

Area of Science:

• Urology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Detrusor overactivity (DO) diagnosis relies on interpreting urodynamic study (UDS) curves, often requiring subjective visual assessment.
• Developing objective, automated diagnostic tools can improve accuracy and efficiency in identifying DO.

Purpose of the Study:

• To create two deep learning-based intelligent diagnosis models for detrusor overactivity (DO).
• To reduce physician dependence on subjective visual analysis of urodynamic study (UDS) curves for DO diagnosis.

Main Methods:

• Convolutional neural network (CNN) models were developed using UDS data from 44 patients (146 DO events, 1863 non-DO events).
• A threshold screening strategy was employed to identify suspected DO segments in UDS curves.
• Model performance was evaluated on 48 independent patients, comparing against classical machine learning models.

Main Results:

• The CNN models achieved high accuracy in diagnosing DO, with 100% accuracy for patients with DO and 78.12% for patients without DO.
• Cross-validation demonstrated superior training and validation accuracy for the developed CNN models.
• The automated system successfully identified DO events, aiding in patient diagnosis.

Conclusions:

• The developed CNN-based DO diagnostic model demonstrates satisfactory accuracy with current data.
• Increasing the dataset size is expected to further enhance the performance of deep learning models for DO diagnosis.