A novel population‐based input function integral estimation method for Ki parametric imaging with shortened dynamic 18F‐FDG PET scans

Abstract Background Dynamic positron emission tomography (PET) is a powerful tool for clinical tumor diagnosis. However, the conventional dynamic scanning duration takes about 60 , which is inconvenient for patients and limits the widespread application of this technology. Purpose This study aims to develop an innovative method to achieve shortened high‐quality parametric imaging based on the Patlak model. Methods We proposed a population‐based input function integral estimation (PBIF‐IE) method. The core of this method is to construct a linear regression model between the early‐stage integral () and the late‐stage mean () of the image‐derived input functions (IDIFs) in the training dataset. The goal is to estimate in the testing datasets using and the linear regression model when only late‐stage dynamic sequences are available. To verify the effectiveness and stability of the model, we set up three testing datasets, A, B, and C, each with different framing protocols for parametric imaging analysis. Furthermore, to explore the potential of our proposed method in shortening scan duration, we evaluated the parameter results obtained by our proposed method under three scan durations (30, 20, and 10 ) using multiple quantitative metrics, including the peak signal‐to‐noise ratio (PSNR), the structural similarity index (SSIM), and the relative error (RE). Results Through multi‐center data studies, we demonstrate the effectiveness of the PBIF‐IE method. Extensive experimental results demonstrate that the PBIF‐IE method outperforms other methods in both estimation and parametric imaging. The 30 dynamic scanning protocol can obtain parameter images that are highly consistent with the 60 scanning protocol, while the 20 dynamic scanning protocol is sufficient for preliminary tumor localization. Conclusions Based on the experimental results, the PBIF‐IE method outperforms other existing methods for shortened parametric imaging. In future research, we plan to explore how the number of dynamic sequences used in the training dataset affects the model construction. This will help us further optimize the parametric imaging process.