Universität zu Köln

Doering, Elena ORCID: 0000-0002-5221-8576 (2025). Artificial Intelligence and Positron Emission Tomography for the Timely Diagnosis and Prognosis of Alzheimer’s Disease. PhD thesis, Universität zu Köln.

PDF Artificial_Intelligence_and_Neuroimaging_for_the_Early_Diagnosis_of_Alzheimer_s_Disease_revisions2.pdf Download (24MB)

Abstract

Alzheimer’s disease (AD) is the most common neurodegenerative disorder worldwide. It is biologically characterized by the accumulation of protein pathology (β-amyloid plaques and tau tangles) and neurodegeneration. For the first time since its first description by Alois Alzheimer more than 100 years ago, effective anti-amyloid therapies are emerging that might slow down disease progression. To initiate treatment early-on, timely diagnosis and prognosis are crucial. Neuropathologic changes in AD begin up to two decades prior to clinical symptoms of dementia, making biomarkers of brain metabolism and protein accumulation promising indicators for timely identification. Positron emission tomography (PET) can depict these processes in vivo; however, the detection of initial and subtle pathology levels in PET scans is challenging using established methods. Artificial intelligence (AI) offers a set of methods capable of identifying complex patterns within PET data, possibly allowing for more nuanced and personalized approaches. The aim of this dissertation was to investigate the potential of AI models applied to PET data to generate clinically relevant markers of disease progression for the timely diagnosis or prognosis of AD. Using AI, various types of AD progression markers were generated and analyzed, predominantly by applying AI to [18F]FDG PET scans, a marker of brain metabolism. In the first study, we observed that AI models can extract reliable information on β-amyloid from [18F]FDG PET and demographic data in individuals with low genetic AD risk (negative APOE-ϵ4 status). The potential of AI for such cross-modal translation can facilitate the workflow for timely diagnoses by reducing the need for multiple PET assessments. The second study demonstrated that [18F]FDG PETderived brain age gaps (BAGs; difference between age and AI-derived brain age) associated with early cognitive dysfunction, but not AD progression. Conversely, MRI-derived BAGs systematically reflected markers of cognitive performance and protein pathology, and predicted future clinical progression, thus representing a global summary score of deviation from healthy brain aging. Regional deviation from healthy brain aging was investigated in a multi-modal PET study, where we showed that the spatial extent of all three hallmark pathologies uniquely reflects AD progression. Thus, spatial extent markers, which we demonstrated can be estimated using deep learning, might be useful for monitoring AD. Finally, after the second study indicated potential for the prediction of clinical progression, we aimed to assess the potential of AI in predicting biological disease progression. Whole-brain [18F]FDG PET scans could be predicted for up to seven years after initial assessment, allowing to preemptively analyze an individual’s future brain metabolic decline. This dissertation highlighted that AI can yield different types of information from PET scans for timely diagnoses via cross-modal translation and estimation of deviation from healthy aging. In the context of prognoses, AI models can predict clinical and biological disease progression. AI-generated information from PET scans can thus enrich diagnostic and prognostic workflows with various patient data that may not otherwise be available to clinicians. Importantly, different markers could be generated based on broadly available [18F]FDG PET, thereby contributing to making diagnostic workflows more efficient and patient-friendly. Future studies should assess the impact of implementing such models in clinical practice. Collaborative efforts between clinicians and AI researchers could ultimately enable treatment initiation before patients experience debilitating cognitive decline from AD.

Item Type:	Thesis (PhD thesis)
Creators:	Creators Email ORCID ORCID Put Code Doering, Elena elena.doering@uk-koeln.de orcid.org/0000-0002-5221-8576 UNSPECIFIED
URN:	urn:nbn:de:hbz:38-781160
Date:	2025
Language:	English
Faculty:	Faculty of Medicine
Divisions:	Faculty of Medicine > Nuklearmedizin > Klinik und Poliklinik für Nuklearmedizin
Subjects:	Data processing Computer science Natural sciences and mathematics Medical sciences Medicine
Uncontrolled Keywords:	Keywords Language Alzheimer's disease English Artificial intelligence English Positron emission tomography English
Date of oral exam:	11 October 2024
Referee:	Name Academic Title Beyer, Andreas Prof. Dr. Grefkes-Hermann, Christian Prof. Dr. med.
Refereed:	Yes
URI:	http://kups.ub.uni-koeln.de/id/eprint/78116