Engineering competences are also becoming increasingly dominant in medicine

“Universities are responsible for transferring and utilising knowledge, through education, RDI contracts and public outreach. The latter helps build public consensus based on scientific facts, that is this activity helps maintain the soundness of social thinking," said János Levendovszky, the BME’s Vice-Rector for Research and Innovation, in his speech introducing the first act of the second season of the Future Planner series.

This time the public outreach lectures were organised around the theme “Artificial Intelligence in the service of medical science”. “AI as a disruptive technology has already brought about very rapid transformations, and it still has huge potential, so there is an understandable social need for AI to revolutionise medicine,” said János Levendovszky. He added that this shows that technical and engineering competences are becoming more and more dominant in medicine, and that the BME is capable of conducting exploratory research and developing solutions to fix the problems of application.

Péter Antal, Associate Professor of the Department of Artificial Intelligence and Systems Engineering at the Faculty of Electrical Engineering and Informatics, delivered a lecture entitled “Research into the Medical Applications of Artificial Intelligence”. He briefly described the types of AI, from artificial narrow intelligence (ANI) to deep learning systems, to explainable, trusted, human-centric and as-yet-unavailable human-level AI.

The following situations in healthcare can be considered for application: diagnosis (patterns and predictions), treatment (interventions and outcomes), decision-making (uncertainty and utility), and explanation (alternatives). The current models are not suitable for decision support by themselves, but with the right tools and frameworks they have the potential for strong performance,” explained Péter Antal. He also reported on a BME research project to improve the prediction of chronic diseases using the UK Biobank database.

The next two speakers were the first-prize winners of a joint article competition of the BME, the journal Élet és Tudomány and the Pro Progressio Foundation. Awarded in the lecturer category, Department Engineer Attila Zoltán Jenei spoke about the potential use of AI in combating dementia. The disease already affects 50 million people and is projected to reach 113 million by 2050. The aim is to detect it as early as possible, the most common tools being MRI and CT, but the department is working on a method that does not require any invasive procedure: speech-based early detection.

Results show that speech is an important biomarker as it can indicate early-stage dementia with 85-90% accuracy (77-85% in non-laboratory settings). Future research may include the augmentation of the samples (better generalisation and classification) and the creation of a dedicated database for the Hungarian language, said Attila Zoltán Jenei.

Doctoral student Szabolcs Torma, the first-prize winner in the student category of the competition, presented his paper entitled “Artificial Intelligence in Neuroimaging”. What justifies the technology in this field is that fMRI is expensive, offers limited spatial access and often provides only low resolution imaging for the studying brain functions. Generative AI offers new solutions: it creates realistic synthetic images, improves image quality, and helps unify data.

To do this, AI uses methods such as data augmentation (adding synthetic data to improve the accuracy and sensitivity of classification models), data harmonisation (transforming heterogeneous data into a uniform format, preserving biological patterns) and super-resolution (reconstructing high-resolution images from low-resolution inputs, improving diagnostic accuracy). For the time being, the high computational and energy demand and the need for further model developments present a major challenge, but there is still great potential for increasing efficiency and generating complex, four-dimensional, dynamic fMRI data more accurately, as Szabolcs Torma pointed out.

Ákos Gózon, editor-in-chief of Élet és Tudomány, moderated the discussion, during which the speakers answered several questions. For example, it was discussed how AI misrepresentations are a big problem. Even though Péter Antal acknowledged that in some areas AI performs better, these misrepresentations still indicate that it is still far from the level of human rationality. Attila Zoltán Jenei was asked, among other things, whether it is possible to increase accuracy by longitudinal analysis. The researcher replied that in theory it certainly is possible, but in practice it is difficult because we do not know in advance who will develop dementia.

Zoltán Jenei Attila , Szabolcs Torma, Péter Antal, Ákos Gózon

Rector's Office, Directorate of Communications