Texture features-guided image reconstruction kernel method forF-FDG delayed PET imaging.

Positron emission tomography (PET) is a crucial imaging technology that is widely used for cancer staging and treatment response assessment. Delayed PET imaging (or dual-time-point imaging) involves a second PET acquisition step approximately 33 min after the initial scan in this study, providing more dynamic biological information, improving lesion detectability, and allowing patients to rest during the scan. However, with increasing time, the number of photons in the second scan decreases, which raises the difficulty of the second PET image reconstruction process. Due to the extended time between the tracer injection step and the second PET scan, the photon count during the second scan decreases, leading to difficulties when reconstructing the second PET image. To effectively reconstruct the second image, we propose a new reconstruction algorithm that utilizes texture features from the first PET image to assist in reconstructing the second PET image.In this study, to effectively reconstruct delayed-scan PET images, we propose a novel reconstruction method. This method introduces texture features from the first PET image to assist in the process of reconstructing the second PET image, thereby enabling the PET images to more effectively retain their clinical significance. We extract texture features using the gray level co-occurrence matrix, then combine these features with additional gray-level characteristics to form a new feature vector, which is subsequently incorporated into the kernel-based reconstruction method, enhancing the reconstruction process and improving the quality of the delayed PET image reconstruction. We used the peak signal-to-noise ratio, the mean absolute error and structural similarity index measure (SSIM) as image quality assessment metrics and compared our method with other existing reconstruction methods. In addition, we conducted a more detailed comparison across regions of interest (ROIs).Our experiments were conducted with data acquired from 32 real patients. Compared with other competing methods, our approach achieved a certain level of improvement, with a PSNR value of 32.53 dB and an SSIM of 0.904. Relative to those of the maximum likelihood expectation maximization method, these metrics improved by 13% and 7%, respectively. Within the ROIs, our method also showed closer agreement with the ground truth, preserving the highly metabolic regions in the images.Our method represents a novel application of reconstruction techniques to delayed imaging, incorporating heterogeneous texture features from prior images for the first time. This approach significantly advances the field of medical image processing by improving the reconstruction quality of delayed PET images, providing clinicians with more reliable and detailed information for patient care. Clinically, the texture features in delayed imaging provide structural information that eliminates the need for CT attenuation correction during delayed PET scans, thereby reducing patient radiation exposure and minimizing the risks associated with repeated scanning.