Artificial intelligence in biomedicine to make research advance faster
- The future of biomedicine is already being written in algorithms
- More accurate medical decisions
- Mistrust and biases, obstacles to overcome
In recent years, artificial intelligence (AI) and its application and integration in different areas has become increasingly tangible in our daily lives. In the field of biomedicine and health, the scenarios that arise are encouraging: perfecting the early detection of any pathology, obtaining more precise diagnoses or even monitoring treatments in real time. However, any light emanating from a new technology generates shadows that need to be dispelled.
The qualitative leap that AI is causing in biomedicine has three main areas: research, clinical practice and training. In all of them, new initiatives are already being developed that will lead to improvements for both medical professionals and patients. However, while all this revolution is taking place, there are still many questions to be answered regarding its clinical application in real-life settings, which has not yet reached the same dizzying pace of development.
The future of biomedicine is already being written in algorithms
"Everything we do at the Barcelona Supercomputing Center, as in the whole practice of biology and biomedicine, is increasingly directly related to AI". Dr. Alfonso Valencia, ICREA professor and director of the Department of Life Sciences at the Barcelona Supercomputing Center (BSC-CNS), is one of the leading figures in the field of bioinformatics and the application of AI to the resolution of biomedical problems. Valencia participates in a project to identify new biomarkers to predict the risk of relapse in acute lymphoblastic leukaemia supported by the CaixaResearch call for research in health.
From his perspective as director of the Department of Life Sciences at the Barcelona Supercomputing Center (BSC-CNS), he confirms that AI is already a fundamental element in the daily work of his team: "We are developing initiatives related to the interpretation of medical data, such as foundational language models trained in Spanish and in the co-official languages to apply them to biomedical texts with the aim of interpreting and understanding medical reports. We are also using it to understand and extract information from X-rays and medical images. But we are also applying AI to genomes and proteins to investigate how it can help design new drugs".
Another major area of research being carried out by Dr Alfonso Valencia's team is the development of so-called digital twins: "Digital twins are engineering constructs that simulate in real time the functioning of a system. A good example is car factories, where a digital twin can replicate any circumstance without having to modify anything in the real world. At the BSC we are simulating atomic systems, proteins and drugs; we also simulate cell behaviour to analyse how a tumour evolves and responds to a drug, how it interacts with the environment or with the immune system... All these examples complement traditional clinical trial systems because we are still a long way from being able to simulate the full complexity of an organism and the human body".
More accurate medical decisions
The precision that AI can bring to healthcare personnel in decision-making is one of the most interesting applications. For example, diagnostic imaging is already a tangible reality, as Alfonso Valencia explains: "The most extensive study to date on the use of AI in the analysis of X-rays has recently been published. The study's findings show that AI systems are able to detect certain features in X-ray images more accurately than medical professionals. However, interaction with the patient in a real-world environment still requires further development".
This diagnostic aid is essential in emergency situations where fast action is needed. A great example is the project led by Dr. Natalia Pérez de la Ossa, coordinator of the Stroke Unit at the Germans Trias i Pujol Hospital and researcher at the Germans Trias i Pujol Research Institute (IGTP), which has the support of the CaixaImpulse Health Innovation 2023 call for proposals. Its aim is to rapidly categorise stroke patients using AI algorithms, which could improve their chances of recovery by 10%. "We are designing a tool that can be used in ambulances to be able to predict the type of stroke a person is suffering with the data collected at the time of the first pre-hospital care, before diagnostic tests or brain imaging, and decide to which hospital it is best to transfer them, as the appropriate therapy in each case can vary a lot," he says.
Another big area where we are seeing advances is real-time monitoring of patients' health status after diagnosis. Pérez de la Ossa explains its potential application in the monitoring of neurodegenerative diseases, for example. Using data collected through existing devices, new AI-based support tools can identify early signs of risk: "Falls, changes in tone of voice, etc., in Parkinson's patients, for example, can indicate possible fluctuations, and all this is important because it will allow healthcare professionals to act quickly and anticipate possible complications".
For his part, Dr Rodrigo Menchaca, founding partner and director of AIS Channel, and director of Digital Skin at ISDIN, alludes to the help that AI provides in the area of training: "Today we can be assisted by a surgical navigation system with AI that guides us in surgery. That is, by means of a system that has been trained to learn a surgery and all the surgical phases, we manage to elaborate a surgical GPS, a heat map that shows the next two ideal steps of the surgery".
Mistrust and biases, obstacles to overcome
One of the biggest risks of applying AI in biomedicine is that biased algorithms can lead to unfair medical decisions or misdiagnosis. These biases are an intrinsic problem in the development of any AI tool and can be acquired at any point in the process: from the selection of the data with which the systems are trained to the interpretation of the results by professionals. According to Alfonso Valencia, "the solution is for the systems to incorporate a margin of error when reporting any result". In addition, the professional must understand that the results are only predictions. And the good news is that the AI industry recognises the need to include these margins and there is a unified response to drive the development of tools that control bias at all levels and monitor system protocols to ensure reliability.
The definitive implementation of AI will also have to overcome some reluctance among healthcare professionals due to fear of the unknown, of not knowing how to interpret the information or of not understanding the tool properly. For Dr Pérez de la Ossa, these doubts could be dispelled with appropriate training so that medical teams can see the value that AI can bring to their daily work, for example, by freeing up their time from more repetitive management tasks so that they can devote it to treating patients. Moreover, as Pérez de la Ossa adds, technology can give us information that we would not otherwise have: "For example, from an electrocardiogram, an AI system can tell us if that patient is at risk of suffering some kind of arrhythmia. That is something that we, with our eyes, cannot see. It is in cases like this that AI has a real added value and we want to incorporate it into our daily practice".
All these conclusions were the protagonists of the CaixaResearch debate "Artificial intelligence in biomedicine: present and future", which was broadcast on 22 March and whose recording can be consulted via the link above. As the experts noted, AI is revolutionising the field of biomedicine by enabling more precise decision-making, reducing the margin of error and implementing advances more quickly.