Deep learning-based segmentation of Jaszczak ACR phantom images for optimized Radium-223 dosimetry

Abstract

Precise and personalized absorbed dose estimation in radionuclide therapy is crucial for optimizing treatment efficiency while minimizing harm to healthy tissues. Radium-223 dichloride (Ra-223), an alpha emitter used in treating metastatic castration-resistant prostate cancer, has shown positive results in extending patient survival. However, the current practice of uniform Ra-223 activity administration based solely on patient weight can lead to suboptimal treatment outcomes. Treatment efficacy evaluation involves quantifying activity and absorbed dose through image quality analysis, revealing potential areas for optimization. This work introduces an innovative approach that integrates a deep learning-based model for automated segmentation of the Jaszczak ACR phantom—a tool for image quality analysis in nuclear medicine—with Monte Carlo simulation for dosimetry. The model exhibits efficient segmentation, surpassing 83.7 % in class-wise Dice coefficients, offering a timeefficient alternative to manual segmentation. The study highlights the superior performance of the 89 keV energy window in image quality parameters, emphasizing its role in lesion detection. Additionally, it addresses challenges in achieving accurate quantitative outcomes in nuclear medicine applications, particularly in Ra-223 therapy. These insights contribute to refining dosimetry protocols for Ra-223, enhancing the precision of quantitative outcomes in nuclear medicine. The practical implications extend to improving daily routines for clinical professionals in nuclear medicine applications, showcasing the potential of advanced imaging techniques and computational tools in optimizing Ra-223 therapy.