Background <p>In dynamic PET with tracer kinetic modeling, model complexity is an important but often under-recognised challenge affecting robust parameter estimation, particularly for noisy data. Traditional methods often neglect tissue heterogeneity and apply a single model universally. We applied a model selection approach alongside delay and motion correction, enabling the selection of models with varying complexity to better account for tissue heterogeneity.</p> Results <p>The study included five subjects with breast cancer undergoing dynamic ¹⁸F-FDG PET imaging using a long axial field of view scanner. Voxel-wise kinetic model parameter estimation utilized five compartmental models, with the best model chosen using the Akaike Information Criterion. The model selection revealed diverse kinetic models within breast cancer lesions voxel-wise, with reduced parameter estimation variability attributed to the choice of simpler models. Applying delay and motion correction reduced the mean coefficient of variation in estimated kinetic parameters by 25%.</p> Conclusions <p>We applied a standard model selection approach to identify the optimal compartmental model for voxel-wise parameter estimation in long field-of-view dynamic PET imaging. Our results demonstrate that accounting for tissue heterogeneity in breast lesions is critical for accurate quantification. Additionally, delay and motion correction were shown to improve image quality, enhance quantification accuracy, and support more reliable model selection.</p> Clinical trial registration <p>Clinical trial number: not applicable.</p>