CLASP
The Centre for Linguistic Theory and Studies in Probability

From CT to CNN’s: Deep Learning for Neuroimaging-Based Diagnosis of Alzheimer’s and Dementia

Abstract

Alzheimer’s disease, a leading cause of dementia, presents complex diagnostic challenges due to its gradual progression and overlapping symptoms with other neurological disorders. While current clinical diagnosis uses a combination of plasma and CSF biomarkers, genetic factors (such as links to chromosome 21), and advanced neuroimaging methods like MRI and PET to detect amyloid and tau accumulation, these tools can be costly and time-intensive. CT remains a common first-line modality in hospitals for broader neurological assessments like trauma or hydrocephalus. In this framework, deep learning can act as a diagnostic tool—offering automated, scalable, and highly accurate analysis of neuroimaging data. By training models like nnUNet and MedNeXT on diverse datasets including H70, NUS, TBI, and pediatric brain scans, we can achieve clinically validated segmentation of brain tissue and pathologies. This presentation discusses the training, validation, and inference workflows of advanced deep learning models, emphasizing how DL is faster, more consistent, and can be used for earlier detection of neurological disorders.