•Notícia

The Artificial Intelligence Decision Tools for Tumour Diagnosis (AIDTumour) project is being funded by the Interministerial Commission for Science and Technology (CICYT) and directed by Alfredo Vellido, a member of the UPC Soft Computing (SOCO) research group. Vellido is coordinating a team of 14 researchers in creating a functional prototype that will assist doctors in improving diagnosis in clinical oncology, in particular of brain tumors, based on magnetic resonance spectroscopy data. This is a cutting-edge initiative in the field of medical decision-making support systems, an area in which Spain has been lagging behind the rest of Europe. 

In addition to the SOCO research group of the UPC’s Department of Software, the team is made up of scientists from Liverpool John Moores University, the University of Manchester, and the NHS Clatterbridge Centre for Oncology in Bebington in the United Kingdom (a leading country in research into oncology decision-making support systems), as well as the Integrated Reasoning Group of the National Research Council Canada. The Group for Biomedical Applications of Nuclear Magnetic Resonance Spectroscopy (GABRMN) of the Universitat Autònoma de Barcelona has collaborated on the project by providing the medical data. 

Companies and hospitals, including Aleasoft, Microart, Vall d'Hebron Hospital, Bellvitge University Hospital, Sant Joan de Déu Hospital, Barcelona Biomedical Research Park, and Cetir Grup Mèdic, have expressed an interest in the project, both in the development of its theoretical aspects and in the possibility of implementing the new system to support clinical diagnosis.

The scientists participating in the AIDTumour project are seeking to create technology that supports the diagnosis of brain tumors. Thus they do not intend to create a decision-making system but rather one that would assist the specialist and become a kind of “second opinion” in the medical diagnosis, which is particularly important in the field of oncology.

The type of brain tumor investigated in the project is not easy to diagnose with certainty. In fact, the best way to be certain of the diagnosis of a tumor is to perform a biopsy but unfortunately this is not advisable where a brain tumor is concerned. In this area, doctors and radiologists have to work with non-invasive techniques and this creates a series of restrictions.

The multi-centre data used on the project are spectra obtained using nuclear magnetic resonance. Some of the frequencies of these spectra may be related to the presence of different chemical compounds in a tumor, which serve as indicators. Based on the observation of these indicators, the few doctors who are expert in the interpretation of these data are usually capable of recognizing each tumor. In contrast, a machine is capable of working with data populations and making use of much richer information.

In fact, the system being developed by the SOCO group is especially useful in ambiguous cases. When anomalies are present, these can be due to an atypical tumor or simply due to measurement problems. It has been shown that these atypical cases can be better characterized and visualized with the new diagnosis support system, since the machine can detect an anomalous case by comparing it with other anomalous cases and then propose hypotheses that could explain the anomaly. This can therefore assist doctors in making informed decisions in clinical diagnosis.

Currently the project is in the final stage of prototype design and the SOCO group is focusing its efforts on integrating these tools into a system of decision support with advanced visualization techniques. The aim is to make the system more advanced, intuitive, and comfortable to use. A prototype is being designed to enable subsequent manufacture of an effective tool to support medical specialists and it will be used this year on an experimental basis at the NHS Clatterbridge Centre for Oncology in Bebington, one of the British participants in the project.

The results of the AIDTumour project research have been recently disseminated through various conferences and international journals including the publications Neural Networks, Neurocomputing, Computers in Biology and Medicine and Biomedical Signal Processing & Control.

The term “soft computing” was coined in the mid-nineties to describe the combined use of different computing techniques that have been developed over the last thirty years, including fuzzy systems, neural networks and evolutionary algorithms.

What these methodologies have in common is that they abandon binary logic, static analytical models, rigid classifications, and determinist searches in approaching real-world problems that are incompletely or badly defined and therefore difficult to model. In these cases, precise models may simply not exist or they may be too impractical or costly to implement. Thus it is necessary to use approximate reasoning systems capable of processing such far-from-perfect information in a flexible way. These types of systems and the methodologies they are based upon with fall within the field of soft computing.