Towards Artificial Intelligence that Healthcare Personnel Can Understand
The doctoral thesis of Miquel Miró Nicolau has developed a set of methods to identify the quality of explanations and increase the reliability of predictions made by artificial intelligence systems used in the healthcare sector.
In recent years, the use of neural networks has experienced significant growth, impacting various sectors from technology to medicine. These computational structures, inspired by the functioning of the human brain, have shown great capacity for managing large amounts of information. However, the reliability and explainability of the predictions made by these artificial intelligence systems continue to be major issues. The lack of transparency in their functioning and the difficulty in understanding how they reach their conclusions generate distrust and concerns about their application in critical environments, such as healthcare.
Miquel Miró Nicolau’s doctoral thesis, recently defended at the University of the Balearic Islands, aims to provide solutions to the problem of explainable artificial intelligence systems, especially in their potential applications in the medical field. To achieve this, the researcher uses a set of techniques known as explainable artificial intelligence (XAI), which seek to make the decision-making processes of artificial intelligence systems understandable and transparent.
The researcher has developed various methods to objectively identify the quality of the explanations provided by artificial intelligence. To do this, he has created a dataset and mathematical measures that allow the identification of erroneous explanations, thus enabling their future exclusion. This approach helps users to have more confidence in the reliability and explainability of artificial intelligence systems.
The methods and techniques developed as part of his doctoral thesis have been tested in collaboration with Son Espases University Hospital (Palma) to assess whether their use can help radiologists perform their tasks more efficiently. They have also been used in a project in collaboration with researchers from the University of Normandy (France), focusing on the use of digital twins—digital copies of a human body—to analyze the application of these techniques for detecting brain tumors through magnetic resonance imaging.
Miquel Nicolau’s research has been conducted as part of his research activity as a member of the Graphics and Computer Vision and AI Research Unit and the Artificial Intelligence Applications Laboratory (LAIA) at the UIB.
Doctoral Candidate: Miquel Miró Nicolau. Title: XAI 4 MIA - Explainable Artificial Intelligence for Medical Image Analysis. Thesis Supervisors: Dr. Gabriel Moyà Alcover and Dr. Antoni Jaume Capó. PhD Program in Information and Communication Technologies