An Investigation on Cross-Tracer Generalizability of Deep Learning-Based PET Attenuation Correction

Attenuation correction (AC) is a critical step to ensure accurate quantitative PET imaging. To eliminate the radiation dose from CT, deep learning (DL)-based methods have been extensively investigated to generate the CT-equivalent attenuation map (<i>μ</i>-CT) directly from the PET signal. However, almost all previous studies only focus on ¹⁸F-FDG due to its extensive data availability which is suitable for DL model training. For other less common tracer types, it is generally believed that new models must be trained separately on these tracer-specific data to ensure reasonable performance. In this work, we explored the cross-tracer generalizability of <i>μ</i>-DL generation DL models - primarily focusing on whether a model trained on a commonly used tracer like ¹⁸F-FDG can be effectively applied to less common tracers such as ⁶⁸Ga-DOTATE and ¹⁸F-Fluciclovine, and vice versa. Unlike methods that directly generate attenuation-corrected (AC) PET images from non-attenuation corrected (NAC) PET images or MLAA reconstructions, we generate the CT-based deep learning attenuation maps (<i>μ</i>-DL) using MLAA reconstruction with the combined input of attenuation maps (<i>μ</i>-MLAA) and tracer activity (<i>λ</i>-MLAA). This <i>μ</i>-DL is then used for attenuation correction to obtain the final AC PET image. Our comprehensive evaluations on both <i>μ</i>-CT generation and the PET reconstruction found that the DL model trained on one specific tracer can be adapted to other tracers with competitive performance when compared to the tracer-specific trained DL model. The ¹⁸F-FDG-trained model demonstrated the best generalizability when applied to less common tracer types which often have relatively fewer available data for training. Additionally, we investigated the role of the <i>μ</i>-MLAA and <i>λ</i>-MLAA as inputs for the network performance. We found that combining both inputs resulted in the best performance, but the <i>μ</i>-MLAA contributed more significantly compared to the <i>λ</i>-MLAA.