Development of Personalized Dosimetry Models Using Radiomics and Deep Learning in Theranostic Nuclear Medicine

Theranostics is an emerging sector in nuclear medicine focused on personalized treatment to enhance patient outcomes. Dosimetry is crucial for planning, utilizing calculated absorbed dose distributions to refine injected activities. Yet, clinical implementation is hindered by the necessity for multiple imaging points, lengthy post-therapy acquisitions, and intricate data analysis. Radiomics and AI can streamline dosimetry workflows with innovative multitasking models.

Radiomics is a technique that extracts novel quantitative features from medical images, widely used in classification, prognosis, or prediction in oncology. In contrast, deep learning is a subset of machine learning that uses neural networks to create decision models from data, encompassing classification, detection, and segmentation applications. The integration of radiomics and deep learning in theranostic dosimetry can automate timepoint selection, predict tracer kinetics, and ultimately translate absorbed dose distributions from the internal dosimetry reference standard.