Artificial Intelligence Research Group
Multidisciplinary R&D unit that coordinates the design, validation and integration of artificial intelligence-based systems developed to support clinical decision processes
Clinical decision support systems can be thought of as digital assistants of modern medicine. We produce solutions that increase patient safety by analyzing electronic health records, laboratory data, vital signs and clinical notes.
Our systems are designed as "assistant algorithms" that help physicians but do not take over decision authority. This approach supports human expertise while ensuring that clinical responsibility remains with the physician.
The purpose of our group is to develop artificial intelligence-supported systems to support diagnosis-treatment decisions and optimize clinical workflows. These systems increase patient safety while accelerating and improving physicians' decision-making processes.
Interactive flow showing how our AI-supported clinical decision process works
EHR, laboratory, vital signs, clinical notes
Machine learning and pattern recognition
Early warning and risk scoring
Physician assessment and final decision
Critical medical areas where we develop clinical decision support systems
Early detection of sepsis, organ failure and critical conditions
Patient condition stratification and risk assessment
Patient prioritization and urgency determination
Side effect prediction and safety warning systems
Automatic information extraction and document analysis
Resource optimization and length of stay prediction
Our AI-supported decision support systems actively used in clinical settings
Calculates postoperative complication risk based on patient age, comorbidities, planned surgical procedure and preoperative findings, and suggests risk reduction strategies.
Analyzes vital signs, laboratory values and clinical notes to calculate sepsis development risk in real time and provides early warning with high accuracy.
Predicts acute kidney injury risk with high sensitivity by analyzing creatinine changes, drug use and comorbidity information.
Analyzes potential interactions between patient history, current medications and new prescription recommendations, provides safety warnings and suggests alternative treatment options.
We collaborate with experts from health and technology fields to develop clinical decision support systems
Key research topics we focus on to improve the reliability and effectiveness of clinical decision support systems
We make the decision-making processes of algorithms transparent. Clinicians should be able to understand why the system makes a particular recommendation and evaluate the rationale behind this recommendation.
We bridge structured data (laboratory, vital signs) with unstructured data (doctor notes, discharge summaries). We extract meaningful information from clinical texts using natural language processing.
We predict future risks by monitoring changes in the patient's condition. This approach is critically important especially in intensive care and emergency medicine applications.
We ensure that developed systems are tested in real clinical environments and feedback from health professionals is received. Continuous monitoring and improvement of system performance is targeted.
We develop ethical frameworks to ensure that systems protect patient privacy, minimize bias, and support but not replace clinical decisions.
We provide decision support without disrupting clinical workflows by ensuring seamless integration with existing hospital information systems. We target user-friendly interfaces and minimal learning curve.
Example of interactive dashboard used in clinical settings
Real-time risk analysis and alert system
Concrete impacts of our clinical decision support systems on patient safety and clinical outcomes
Would you like to collaborate with us to develop clinical decision support systems?
System testing and validation in real environments with hospitals
Model development and improvement with anonymous clinical data
Productization and commercialization of developed algorithms