Neuro-fuzzy modeling: an accurate and interpretable method for predicting bladder cancer progression

Area of Science:

• Oncology
• Artificial Intelligence
• Biomarkers

Background:

• Traditional statistical methods offer limited accuracy in predicting cancer progression.
• Accurate cancer progression prediction is crucial for personalized treatment strategies.
• Artificial Neural Networks (ANN) show promise but suffer from a lack of transparency ('black box' problem).

Purpose of the Study:

• To develop and compare novel artificial intelligence methods for improved cancer progression prediction.
• To evaluate Neuro-fuzzy modeling (NFM) as a transparent alternative to ANN.
• To identify key clinicopathological and molecular predictors of cancer progression.

Main Methods:

• Combined clinicopathological (stage, grade, age, gender, smoking) and molecular (p53 expression, DNA methylation) data from 117 patients.
• Developed predictive models using Neuro-fuzzy modeling (NFM), Artificial Neural Networks (ANN), and Logistic Regression (LR).
• Assessed model performance using sensitivity, specificity, and accuracy metrics.

Main Results:

• NFM achieved superior prediction accuracy (94-100%) compared to ANN (89-90%) and LR (47-72%).
• NFM demonstrated higher sensitivity (88-100%) and specificity (97-100%) than ANN and LR.
• NFM identified key predictive parameters including age, grade, stage, smoking status, and methylation.

Conclusions:

• Intelligent systems and molecular biomarkers significantly enhance cancer progression prediction accuracy.
• Neuro-fuzzy modeling (NFM) offers a more accurate, sensitive, specific, and transparent approach than ANN.
• Further validation of NFM for routine clinical integration is warranted.