News,
In a recent study published in Scientific Reports, researchers in Egypt trained a deep learning model to predict brain age. They applied a spatial and temporal analysis on MRI datasets to map anatomical features and how they change over time. This model, they report, outperforms others in error margin, achieving a lower Mean Absolute Error (MAE), a measurement of deviations between the predicted and actual age.
Training the model required thousands of MRI scans from individuals across several age groups. They found that the predicted ages accurately matched with actual ages of the individuals. However, accuracy drops among individuals in advanced age groups, gradually increasing from early adulthood. Overall, the accuracy is higher than previous models applied on similar datasets, as measured by the MAE, 3.15 years.
The potential use of this model is early detection and monitoring of the progression of neurodegenerative diseases, such as different forms of dementia. Furthermore, supplemental monitoring of rehabilitation or to track responsiveness to various kinds of interventions, such as diet, drugs or physical activity.
The limitations, however, as the authors themselves report, are the reliance on MRI data processed through anatomical mapping. This method is anatomically unspecific, meaning detailed data might be lost during preprocessing. Furthermore, the application in clinical settings requires ethical considerations, such as data privacy and informed consent.
Potentially, the model can be used for other screening purposes with questionable ethical implications if not regulated. However, regulatory bodies struggle to keep up with technological development. Developers and researchers need to develop methodologies and technology that appleis ethical frameworks to mitigate risks. While fostering a climate of ethical considerations on new technology, we should not restrict innovation, a difficult but necessary balance.
Declaration of AI use: The author used Lumo AI for spelling and grammar-checking during the preparation of this article.
References
Mahmoud, E., Elshennawy, N.M. & Elkholy, A. Hybrid deep learning model for brain age prediction using time-distributed convolutional and bidirectional LSTM networks. Sci Rep 16, 18332 (2026). https://doi.org/10.1038/s41598-026-54198-5
Leave a Reply