Attention-enabled Explainable AI for Bladder Cancer Recurrence Prediction.

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Informatics

Background:

• Non-muscle-invasive bladder cancer (NMIBC) has high recurrence rates (70-80%), leading to repeated procedures and increased costs.
• Current prediction tools for NMIBC recurrence are often inaccurate and lack personalization.
• Effective prediction is crucial for managing patient care and healthcare resource allocation.

Purpose of the Study:

• To develop an interpretable deep learning framework to improve the prediction of NMIBC recurrence.
• To enhance prediction performance by integrating vector embeddings and attention mechanisms.
• To provide patient-specific insights into recurrence risk factors.

Main Methods:

• Developed a deep learning framework incorporating vector embeddings for categorical variables (e.g., smoking status, intravesical treatments).
• Utilized attention mechanisms to identify influential features for personalized risk assessment.
• Evaluated model performance using tabular data, comparing it against conventional statistical methods.

Main Results:

• Achieved 70% accuracy in predicting NMIBC recurrence, outperforming traditional statistical models.
• The model provides clinician-friendly, patient-level explanations of recurrence risk through feature attention.
• Identified novel recurrence factors, including surgical duration and hospital stay, not previously considered.

Conclusions:

• The proposed interpretable deep learning framework significantly improves NMIBC recurrence prediction.
• The model offers valuable patient-specific insights, aiding clinical decision-making and personalized management.
• This approach advances NMIBC risk assessment by incorporating previously overlooked factors and enhancing model transparency.