Artificial intelligence will improve early detection of dementia

Alzheimer’s disease is the most common cause of dementia, affecting millions of older people worldwide. In Spain alone, it is estimated that more than 900,000 people suffer from the disease, which has become a public health priority for the Ministry of Health.

Early detection is key to improving the quality of life of those affected and their families, but catching it in its early stages is not always easy. To try to improve in this area and achieve better results, researchers from the Polytechnic University of Madrid (UPM) applied machine learning and artificial intelligence (AI) techniques to analyze different medical imaging techniques used to diagnose neurological diseases.

“From a clinical point of view, magnetic resonance imaging (MRI) and positron emission tomography (PET) are two medical imaging techniques used in the diagnosis of this type of disease, as they provide additional information about the anatomy and the disease,” explains Consuelo Gonzalo, a researcher at the UPM Biomedical Technology Centre and one of the authors of this work.

“But unfortunately, these tests are not carried out in sync, which makes it difficult for health workers to integrate them and correctly interpret the results,” he continues.

Solving this problem is the goal set by UPM researchers, and the proposal was to develop a methodology using convolutional neural networks.

This tool is a machine learning method that enhances image analysis and computer vision tasks by allowing you to extract important information from digital images, videos, and other visual inputs, and make measurements based on those inputs.

Methodological innovation

To this end, UPM researchers conducted a systematic analysis of MRI and PET images to assess dementia status using different fusion methods (early, late and intermediate fusion).

They then developed and implemented a solution based entirely on 3D convolutional neural networks that extracted features from the entire brain volume in three dimensions. Once this was done, they applied a learning strategy that could handle a highly imbalanced and incomplete dataset.

Extraction of a section of each patient’s brain. /UPM

“To the best of our knowledge, the proposed methodology represents the first work that provides an analysis of different multimodal deep learning-based fusion methods for dementia severity assessment,” explains the UPM researcher.

“The solutions developed in this work could become a decision support tool of great practical interest to neuroscientists,” he adds.

The work, published in the journal Artificial Intelligence in Medicine, also involved the Italian universities of Naples and Rome and Umeå University in Sweden. In future studies, the researchers intend to continue to study the fusion of different modalities, further analyzing the representation properties of common features.

“Approaches aimed at improving the integration of heterogeneous data should be explored, generalizing them to case studies with more than two visualization methods. Attention should also be paid to the explainability of the implemented models, evaluating the decisions made by the networks in comparison with the clinical diagnosis,” they conclude.

Link:

Michela Gravina et al., “A 3D multi-input, multi-output CNN for dementia severity assessment with incomplete multimodal data.” Artificial Intelligence in Medicine (2024).