AI-Based Clinical Rule Discovery for NMIBC Recurrence through Tsetlin Machines

2025-07-26

AI Summary

Current EORTC risk tables exhibit low accuracy—particularly for intermediate-risk non-muscle-invasive bladder cancer (NMIBC) patients—and poor clinical applicability, contributing to recurrence rates as high as 70% and unnecessary overtreatment. To address this, we introduce the first application of the interpretable artificial intelligence model, the Tsetlin Machine (TM), to NMIBC recurrence prediction. Leveraging data from the PHOTO trial, our approach integrates key clinical features—including tumor count, surgeon experience, and length of hospital stay—to construct a high-accuracy, fully transparent, rule-based predictive model. The TM achieves an F1-score of 0.80, significantly outperforming XGBoost, logistic regression, and the conventional EORTC risk tables. Crucially, its inherent interpretability enables clinicians to understand and trust individual predictions, facilitating evidence-based, personalized surveillance and intervention decisions. This work bridges the gap between predictive performance and clinical utility in NMIBC management.

Technology Category

• Machine Learning: Transparent, Interpretable, Explainable ML
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Bladder cancer claims one life every 3 minutes worldwide. Most patients are diagnosed with non-muscle-invasive bladder cancer (NMIBC), yet up to 70% recur after treatment, triggering a relentless cycle of surgeries, monitoring, and risk of progression. Clinical tools like the EORTC risk tables are outdated and unreliable - especially for intermediate-risk cases. We propose an interpretable AI model using the Tsetlin Machine (TM), a symbolic learner that outputs transparent, human-readable logic. Tested on the PHOTO trial dataset (n=330), TM achieved an F1-score of 0.80, outperforming XGBoost (0.78), Logistic Regression (0.60), and EORTC (0.42). TM reveals the exact clauses behind each prediction, grounded in clinical features like tumour count, surgeon experience, and hospital stay - offering accuracy and full transparency. This makes TM a powerful, trustworthy decision-support tool ready for real-world adoption.

Problem

• Develops AI model for predicting NMIBC recurrence accurately
• Improves outdated clinical tools like EORTC risk tables
• Ensures interpretability and transparency in AI predictions

Innovation

• Interpretable AI model using Tsetlin Machine
• Symbolic learner with human-readable logic
• Outperforms XGBoost and Logistic Regression