Deep-JASC: joint attenuation and scatter correction in whole-body 18F-FDG PET using a deep residual network

Abstract

Objective We demonstrate the feasibility of direct generation of attenuation and scatter-corrected images from uncorrected images (PET-nonASC) using deep residual networks in whole-body 18 F-FDG PET imaging. Methods Two-and three-dimensional deep residual networks using 2D successive slices (DL-2DS), 3D slices (DL-3DS) and 3D patches (DL-3DP) as input were constructed to perform joint attenuation and scatter correction on uncorrected whole-body images in an end-to-end fashion. We included 1150 clinical whole-body 18 F-FDG PET/CT studies, among which 900, 100 and 150 patients were randomly partitioned into training, validation and independent validation sets, respectively. The images generated by the proposed approach were assessed using various evaluation metrics, including the root-mean-squared-error (RMSE) and absolute relative error (ARE %) using CT-based attenuation and scatter-corrected (CTAC) PET images as reference. PET image quantification variability was also assessed through voxel-wise standardized uptake value (SUV) bias calculation in different regions of the body (head, neck, chest, liver-lung, abdomen and pelvis). Results Our proposed attenuation and scatter correction (Deep-JASC) algorithm provided good image quality, comparable with those produced by CTAC. Across the 150 patients of the independent external validation set, the voxel-wise REs (%) were − 1.72 ± 4.22%, 3.75 ± 6.91% and − 3.08 ± 5.64 for DL-2DS, DL-3DS and DL-3DP, respectively. Overall, the DL-2DS approach led to superior performance compared with the other two 3D approaches. The brain and neck regions had the highest and lowest RMSE values between Deep-JASC and CTAC images, respectively. However, the largest ARE was observed in the chest (15.16 This article is part of the Topical Collection on Advanced Image Analyses (Radiomics and Artificial Intelligence)

Deep learning-assisted ultra-fast/low-dose whole-body PET/CT imaging

“…During recent years, deep learning algorithms were deployed for various medical image analysis tasks, exhibiting superior performance over traditional strategies [4][5][6][7][8][9][10]. Conventional post-reconstruction PET denoising approaches, such as Gaussian, bilateral and non-local mean filtering, are commonly used in clinical and research settings.…”

Applications of artificial intelligence and deep learning in molecular imaging and radiotherapy

“…Overall, deep learning approaches seem to exhibit better (at least comparable) performance for PET quantification compared to existing state-of-the-art approaches in whole body Hwang et al 2019), pelvic (Arabi et al 2018;Torrado-Carvajal et al 2019), and brain imaging (Liu et al 2017;Gong et al 2018;Blanc-Durand et al 2019). These approaches require at least one MR sequence as input for CT synthesis, whereas deep learning-based joint estimation of attenuation and emission images from TOF PET raw data (Hwang et al 2019;Hwang et al 2018) and direct scatter and attenuation correction in the image domain (Yang et al 2019;Bortolin et al 2019;Shiri et al 2020a;, and sinogram domain (Arabi and Zaidi 2020) could possibly obviate the need for any structural/anatomical images. It should be noted that the information provided by CT images is not ideal for PET attenuation correction owing to the discrepancy between photon energies used in CT and PET imaging.…”

The promise of artificial intelligence and deep learning in PET and SPECT imaging

“…4). Shiri et al [93] trained 2D, 3D, and patch-based ResNets on 1000 wholebody 18 F-FDG images and tested the proposed models on unseen 150 subjects. They performed ROI-based and voxel-based assessments and reported a relative error of less than 5%.…”

“…Most deep learning-based ASC studies focused on brain imaging, which is less challenging compared to whole-body imaging where the anatomical structures are more complex with juxtapositions of various tissues having diverse attenuation properties and irregular shapes. There is obviously a need to evaluate these algorithms in more challenging heterogeneous regions, such as the chest and abdomen [93]. Moreover, the majority of these studies were performed using only one radiotracer (mostly 18 F-FDG) which raises questions regarding the generalizability of the models and the need for retraining and reevaluation on other tracers [92].…”