In the future, artificial intelligence will play a significant role in the field of medicine. One area where AI has already shown promise is in diagnostics. For instance, computers can now accurately categorize images to identify pathological changes. However, training AI to analyze the changing conditions of patients and provide treatment suggestions has been more challenging. At TU Wien and the Medical University of Vienna, researchers have achieved this feat by developing an AI that can provide treatment recommendations for patients in intensive care units suffering from sepsis.
Making the Most of Existing Data
Researchers wanted to explore if the massive amount of data collected in intensive care units could be utilized more effectively. “There is an abundance of data constantly being monitored in these units. Our goal was to determine if this data could be better utilized,” says Prof. Clemens Heitzinger from the Institute for Analysis and Scientific Computing at TU Wien. Prof. Heitzinger is also a Co-Director of the “Center for Artificial Intelligence and Machine Learning” (CAIML) at TU Wien.
While medical professionals base their decisions on established rules and parameters, computers can consider a wider range of factors. In some cases, this broader consideration can lead to even better treatment decisions than those made by humans.
The Computer as Decision-Maker
The project utilized a form of machine learning known as reinforcement learning. According to Prof. Heitzinger, reinforcement learning is not just about simple categorization, but rather about understanding the dynamic progression of a patient’s condition. “This deviation from traditional machine learning is an underexplored area in the medical field,” he explains.
In this project, the computer acts as an agent, making decisions on its own. It is rewarded when the patient’s condition improves and punished if there is deterioration or death. The computer program’s objective is to maximize its “reward” by taking appropriate actions. This approach enables the algorithm to generate treatment strategies with a higher probability of success based on extensive medical data.
AI Outperforms Humans
Sepsis is a leading cause of death in intensive care medicine, and early detection and treatment are critical. Prof. Oliver Kimberger from the Medical University of Vienna states, “Few breakthroughs have been made in this field, making it crucial to explore new treatments and approaches. AI, incorporating machine learning models and other technologies, offers opportunities to enhance diagnosis and treatment of sepsis, ultimately improving patient survival rates.”
Analysis reveals that AI algorithms already outperform human decision-making. “In one study, the AI strategy resulted in a 3% increase in the 90-day mortality cure rate, reaching approximately 88%,” says Prof. Heitzinger.
However, it is worth noting that the AI should not replace medical professionals entirely. It can serve as an additional tool at the bedside. Medical staff can consult the AI and compare its suggestions with their own expertise. Moreover, AI systems can be valuable in educational settings.
Addressing Legal Concerns
Prof. Heitzinger highlights the importance of discussing the legal aspects surrounding AI implementation. “An important question arises: Who will be held responsible if the AI makes a mistake? Conversely, what if the AI is correct, but a human chooses a different treatment option, resulting in harm to the patient?” Asks Prof. Heitzinger. He believes that there is an urgent need for a dialogue on the social framework and the establishment of clear legal regulations concerning the use of AI in medicine.