Early estimation of disease risk from magnetic resonance images (MRIs) can help to reduce the clinical burden of incurable neurological disorders such as Alzheimer’s disease and related dementias (ADRD). Traditionally, researchers’ attempts to identify early biomarkers of future ADRD have relied on neuroanatomic measures defined a priori, including brain volume loss and cortical thinning. Such measures have limited utility due to their modest sensitivity and specificity for the prognostication of ADRD. Recent progress in explainable artificial intelligence (XAI) leverages the ability of deep neural networks to find complex patterns of abnormal neuroanatomic aging that are not apparent to humans and that can better predict ADRD morbidity. Because brain aging is lifelong, such abnormal aging trajectories have the advantage of being detectable relatively early in adulthood to mitigate ADRD risk. Our patient-tailored anatomic maps of brain aging highlight differences in neurosenescence according to sex, decadal age group, biometrics, demographics, and cognitive status. These XAI-empowered findings identify, for the first time, the anatomic substrates of complex endophenotypes whose structural bases were previously thought to be undetectable by MRI. In conclusion, XAI holds considerable potential to assist translational neuroscience, to advance basic studies of brain structure/function, and to develop early biomarkers of ADRD risk in aging adults with normal cognition.
SPEAKER BIOGRAPHY
Andrei Irimia, PhD is a visiting associate professor at King’s College London, currently on sabbatical from the Leonard Davis School of Gerontology at the University of Southern California. Dr. Irimia is a biogerontologist and computational neurobiologist studying how (epi)genetic and environmental factors constrain brain aging in health and disease. In collaboration with the ENIGMA Consortium and with other researchers across the world, his team uses explainable artificial intelligence (XAI), omics, and neuroimaging to characterize risk factors for Alzheimer’s disease (AD). These methods are synergized with biometrics, demographics and with large-scale research on pre-industrial populations to build XAI models that forecast AD conversion in aging adults. Such approaches relate AD risk to accelerated aging, neurovascular calcification, industrialization, urbanization, lifestyle and traumatic brain injury.
