FGD-pet görüntüleme kullanarak alzheimer hastaliğinin radyomik analizi

Alzheimer's disease (AD) demands more advanced and personalized diagnostic methods. This dissertation proposes a radiomics-driven framework using FDG-PET to enhance the diagnosis, staging, and risk evaluation of AD. By integrating image-derived features and machine learning, we aim to facilitate early, non-invasive, and individualized assessment. We developed a fully automated brain radiomics platform capable of distinguishing between CN, MCI, and AD individuals. The hippocampus, entorhinal cortex, and amygdala emerged as key discriminative regions. Our simplified FDG-PET model achieved an AUC of 0.853 for predicting amyloid positivity, a critical pathological hallmark of AD, using features from the hippocampus, inferior parietal lobule, and isthmus cingulate. The platform was further extended for ApoE4 genotype prediction, achieving an AUC of 0.945 with features extracted from the hippocampus, amygdala, thalamus, and pars orbitalis. We also investigated hippocampus-amygdala connectivity, identifying robust biomarkers such as Shape Mesh Volume and GLDM Small Dependence Low Gray Level Emphasis (AUC = 0.88). Subregional analysis of hippocampal and amygdaloid structures revealed additional radiomic features, including GLRLM Long Run Emphasis and GLDM Small Dependence Emphasis, which differentiate AD from MCI and CN, indicating early microstructural and metabolic changes prior to visible atrophy. Our findings establish FDG-PET radiomics as a reliable, non-invasive imaging biomarker for diagnosing and monitoring AD. The proposed framework enables risk stratification and supports clinical decision-making, paving the way for preventive and personalized AD management.