Medical recommender systems based on continuous-valued logic and multi-criteria decision operators, using...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Computational Medicine

Background:

• Digital transformation lags in healthcare, burdening physicians with complex treatment decisions.
• Current systems lack individualized patient care, time efficiency, and physician support.
• Recommender systems offer a solution to aid physicians in treatment selection and workload reduction.

Purpose of the Study:

• To develop a novel, transparent, and safe recommender system for personalized medical treatment.
• To leverage continuous-valued logic and multi-criteria decision operators for accurate outcome prediction.
• To enhance physician decision-making and patient management in complex chronic conditions.

Main Methods:

• Developed a Logic-Operator Neuronal Network (LONN) simulating cognitive, fuzzy logical processes.
• Implemented data synthetization techniques to protect patient privacy and ensure data integrity.
• Created a recommender system comparing two models: one trained on optimal outcomes and another on the full patient population.

Main Results:

• Achieved approximately 75% accuracy with LONN models, demonstrating transparency and safety against adversarial attacks.
• Data synthetization resulted in a ~1% error rate for relevant parameters.
• Successfully predicted individualized therapies and expected treatment durations, with optimal therapies matching both model predictions.

Conclusions:

• The developed recommender system simplifies administrative tasks and enhances patient management quality.
• The transparent and safe LONN methodology offers a robust solution for healthcare decision support.
• The system's potential extends to various clinical scenarios, with future development focused on user-specific requirements.