Sudeshna Bora: Explainable AI for accelerated brain aging in Alcohol Misuse Participants in UK Biobank

Alcohol consumption can lead to accelerated brain ageing. In this study, I analyse different alcohol consumption behaviour with an aim to determine if they contribute to brain age acceleration. I further investigate the brain regions affected by this acceleration.

Using the UK Biobank dataset, I trained a model on 2773 healthy participants to predict their chronological age from their neuroimaging data. Both imaging-derived phenotypes (IDPs) and raw 3D MRI data were used as features. The trained model was then used across various alcohol consumption groups to test if excessive alcohol use showed an acceleration in brain ageing. Two explainable AI approaches were used to examine which brain regions were associated with brain age acceleration.

No acceleration in brain age was detected using the model trained on brain IDPs across all test sets. Whereas for the convolutional neural network model (CNN), in two alcohol consumption groups, namely participants with AUD and participants scoring >15 in Alcohol Use Disorders Identification Test (AUDIT) questionnaire, acceleration in brain ageing was detected.

I found that although all brain regions contributed to the prediction of brain age, especially the ventricular region and the cerebellum was highlighted. However, none of the brain regions was significantly different between the test group and the healthy control set.
Acceleration was found in certain groups of alcohol consumption behaviour. However, further visualisation and statistical analysis rendered my findings inconclusive. Despite this, my study may be considered to provide reference information for future clinical studies.