A machine learning algorithm for automatic tumour board recommendations in prostate cancer patients.

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Multidisciplinary tumour boards (MTBs) are crucial for prostate cancer management but are time-intensive.
• Automating MTB recommendations can improve accessibility and efficiency.

Purpose of the Study:

• To evaluate machine learning (ML) algorithms for automating diagnostic and therapeutic recommendations in prostate cancer patient management.
• To assess the performance of Decision Tree, Random Forest, and K-Nearest Neighbours (KNN) algorithms in predicting MTB decisions.

Main Methods:

• A retrospective dataset of 1929 MTB recommendations (2020-2024) was utilized.
• Three ML algorithms were trained to predict recommendations for PSMA-PET, conventional imaging, active surveillance, and local therapy.
• Model performance was evaluated using accuracy, precision, recall, and F1-score.

Main Results:

• The Random Forest model achieved the highest overall accuracy (66.3%).
• High accuracy was observed for local therapy predictions (F1-score: 0.99).
• Performance was lower for less frequent recommendations (PSMA-PET, active surveillance) due to class imbalance and guideline changes.

Conclusions:

• ML holds promise for replicating MTB decision-making patterns in prostate cancer.
• Current models require further optimization for clinical application, serving as a proof-of-concept.
• Future research should focus on multi-institutional data, prospective validation, and adaptation to evolving guidelines.