The overarching goal of this thesis is to improve patient care and safety while maximizing information extraction from images, using advanced image processing and AI, including machine learning and deep learning. Methodological developments target five domains: data acquisition, curation, segmentation, dosimetry, and automated diagnosis/prognosis. Chapters 2 to 5 focus on data acquisition and personalization in CT, applicable to both SPECT/CT and PET/CT. Chapters 6 to 9 address segmentation for image quantification in hybrid imaging. Segmentation is essential for dosimetry and quantification, especially when one modality is compromised. These chapters present state-of-the-art methodologies for segmenting organs in whole-body and cardiac images across CT, PET, and hybrid imaging scenarios. Chapters 10–14 focus on data curation and artifact correction. Misregistration and modality-specific artifacts challenge hybrid imaging. These chapters propose explainable segmentation-based tools for annotating misregistered PET/CT data. Chapters 15 and 16 introduce AI-based voxel-wise dosimetry. Chapter 15 details an internal dosimetry in theranostics using CT and SPECT data, delivering results comparable to Monte Carlo methods but with a much more affordable time. Chapter 16 describes deep learning models for CT dosimetry, adaptable to modern acquisition protocols. Finally, Chapters 17 to 21 explore automated diagnosis and prognosis. Chapter 17 details a deep learning pipeline for detecting and classifying cardiac amyloidosis, validated internally and externally. Chapters 18 to 21 develop machine learning models that use imaging and dosimetry data from the tumors and healthy organs to predict treatment response and survival outcomes in cancer patients.