Light Bladder Net: Non-invasive Bladder Cancer Prediction by Weighted Deep Learning Approaches and Graphical Data...

Area of Science:

• Urology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Bladder cancer (BCa) presents high recurrence rates, necessitating early and precise detection methods.
• Non-invasive diagnostic approaches are crucial for improving patient outcomes and reducing healthcare burdens.

Purpose of the Study:

• To develop a lightweight and rapid deep learning model, Light-Bladder-Net (LBN), for non-invasive bladder cancer detection.
• To enhance the model's generalization and classification accuracy using urine data.

Main Methods:

• Data augmentation techniques including uniform noise addition and feature selection (mRMR, PCA, SVD, t-SNE) were employed.
• Extracted key vectors were integrated into the dataset, and multiple machine learning models were trained.
• Weighted voting was utilized to combine predictions from various models for improved accuracy.

Main Results:

• The developed model achieved a robust performance in bladder cancer detection.
• Key performance metrics included an accuracy of 0.83, sensitivity of 0.85, specificity of 0.80, and precision of 0.81.
• The results demonstrate the model's effectiveness in identifying bladder cancer from conventional urine samples.

Conclusions:

• The non-invasive diagnostic method provides rapid and cost-effective bladder cancer predictions.
• A free online tool is accessible for clinicians and patients for convenient BCa detection using urine samples.
• This approach facilitates early detection and management of bladder cancer.