Résumé : Background: Selective internal radiotherapy based on transarterial radioembolization (TARE) with yttrium-90 (90Y) microspheres is an established treatment for primary or metastatic liver disease. Purpose: The objective of this work is to optimize the dosimetry of patients treated with 90Y TARE, using positron emission tomography (PET) images. Methods: The NEMA 2012 PET phantom was filled with nearly 3.9 GBq of 90Y activity and acquired at days 0, 3, 5, 7, and 9 on a classic time-of-flight PET/computed tomography (CT) scanner (Philips TF64) and on a silicon photomultiplier (SiPM)-based PET/CT scanner (Philips Vereos). Acquisitions were carried on following the guidelines proposed in a previously published multicentric trial and images were reconstructed by varying and combining the available parameters. Comparisons were performed to identify the best set(s) of parameters leading to the most accurate 90Y-PET image(s), in terms of activity distribution. Then, for both scanners, the best images were analyzed with Simplicit90Y, a personalized dosimetry software using multicompartmental Medical Internal Radiation Dose model. The comparison between measured and true doses allowed to identify the image granting the most consistent dose estimations and, therefore, to designate the set of parameters to be applied on patients’ data for the reconstruction of optimized clinical images. Posttreatment dosimetry of four patients was then realized with Simplicit90Y using optimized imaging datasets. Results: Based on activity distribution comparisons and dose estimations over phantom and patients data, the SiPM-based PET/CT system appeared more suitable than the photomultiplier tube-based TF64 for 90Y-PET imaging. With the SiPM-based PET/CT system, reconstructed images with a 2-mm voxel size combined with the application of the point spread function correction led to the most accurate results for quantitative 90Y measures. Conclusions: For the SiPM-based PET/CT scanner, an optimized set of reconstruction parameters has been identified and applied on patients’ data in order to generate the most accurate image to be used for an improved personalized 90Y-PET dosimetry, ensuring a reliable evaluation of the delivered doses.