Deep Generalized Learning Model for PET Image Reconstruction.

Low-count positron emission tomography (PET) imaging is challenging because of the ill-posedness of this inverse problem. Previous studies have demonstrated that deep learning (DL) holds promise for achieving improved low-count PET image quality. However, almost all data-driven DL methods suffer from fine structure degradation and blurring effects after denoising. Incorporating DL into the traditional iterative optimization model can effectively improve its image quality and recover fine structures, but little research has considered the full relaxation of the model, resulting in the performance of this hybrid model not being sufficiently exploited. In this paper, we propose a learning framework that deeply integrates DL and an alternating direction of multipliers method (ADMM)-based iterative optimization model. The innovative feature of this method is that we break the inherent forms of the fidelity operators and use neural networks to process them. The regularization term is deeply generalized. The proposed method is evaluated on simulated data and real data. Both the qualitative and quantitative results show that our proposed neural network method can outperform partial operator expansion-based neural network methods, neural network denoising methods and traditional methods.

Head-to-head intra-individual comparison of total-body 2-[18F]FDG PET/CT and digital PET/CT in patients with malignant tumor: how sensitive could it be?

OBJECTIVES: To comparatively evaluate the lesion-detecting ability of 2-[18F]FDG total-body PET/CT (TB PET/CT) and conventional digital PET/CT. METHODS: This study enrolled 67 patients (median age, 65 years; 24 female and 43 male patients) who underwent a TB PET/CT scan and a conventional digital PET/CT scan after a single 2-[18F]FDG injection (3.7 MBq/kg). Raw PET data for TB PET/CT were acquired over the course of 5 min, and images were reconstructed using data from the first 1, 2, 3, and 4 min and the entire 5 min (G1, G2, G3, G4, and G5, respectively). The conventional digital PET/CT scan acquired in 2-3 min per bed (G0). Two nuclear medicine physicians independently assessed subjective image quality using a 5-point Likert scale and recorded the number of 2-[18F]FDG-avid lesions. RESULTS: A total of 241 lesions (69 primary lesions; 32 liver, lung, and peritoneum metastases; and 140 regional lymph nodes) among 67 patients with various types of cancer were analyzed. The subjective image quality score and SNR (signal-to-noise ratio) increased gradually from G1 to G5, and these values were significantly higher than the values at G0 (all p < 0.05). Compared to conventional PET/CT, G4 and G5 of TB PET/CT detected an additional 15 lesions (2 primary lesions; 5 liver, lung, and peritoneum lesions; and 8 lymph node metastases). CONCLUSION: TB PET/CT was more sensitive than conventional whole-body PET/CT in detecting small (4.3 mm, maximum standardized uptake value (SUVmax) of 1.0) or low-uptake (tumor-to-liver ratio of 1.6, SUVmax of 4.1) lesions. CLINICAL RELEVANCE STATEMENT: This study explored the gain of the image quality and lesion detectability of TB PET/CT, compared to conventional PET/CT, and recommended the appropriate acquisition time for TB PET/CT in clinical practice with an ordinary 2-[18F] FDG dose. KEY POINTS: • TB PET/CT increases the effective sensitivity to approximately 40 times that of conventional PET scanners. • The subjective image quality score and signal-to-noise ratio of TB PET/CT from G1 to G5 were better than those of conventional PET/CT. • 2-[18F]FDG TB PET/CT with a 4-min acquisition time at a regular tracer dose detected an additional 15 lesions compared to conventional PET/CT.

Respiratory-gated PET imaging with reduced acquisition time for suspect malignancies: the first experience in application of total-body PET/CT.

OBJECTIVES: This study aimed to investigate the performance of respiratory-gating imaging with reduced acquisition time using the total-body positron emission tomography/computed tomography (PET/CT) scanner. METHODS: Imaging data of 71 patients with suspect malignancies who underwent total-body 2-[18F]-fluoro-2-deoxy-D-glucose PET/CT for 15 min with respiration recorded were analyzed. For each examination, four reconstructions were performed: Ungated-15, using all coincidences; Ungated-5, using data of the first 5 min; Gated-15 using all coincidences but with respiratory gating; and Gated-6 using data of the first 6 min with respiratory gating. Lesions were quantified and image quality was evaluated; both were compared between the four image sets. RESULTS: A total of 390 lesions were found in the thorax and upper abdomen. Lesion detectability was significantly higher in gated-15 (97.2%) than in ungated-15 (93.6%, p = 0.001) and ungated-5 (92.3%, p = 0.001), but comparable to Gated-6 (95.9%, p = 0.993). A total of 131 lesions were selected for quantitative analyses. Lesions in Gated-15 presented significantly larger standardized uptake values, tumor-to-liver ratio, and tumor-to-blood ratio, but smaller metabolic tumor volume, compared to those in Ungated-15 and Ungated-5 (all p < 0.001). These differences were more obvious in small lesions and in lesions from sites other than mediastinum/retroperitoneum. However, these indices were not significantly different between Gated-15 and Gated-6. Higher, but acceptable, image noise was identified in gated images than in ungated images. CONCLUSIONS: Respiratory-gating imaging with reduced scanning time using the total-body PET/CT scanner is superior to ungated imaging and can be used in the clinic. KEY POINTS: • In PET imaging, respiratory gating can improve lesion presentation and detectability but requires longer imaging time. • This single-center study showed that the total-body PET scanner allows respiratory-gated imaging with reduced and clinically acceptable scanning time.

Dynamic PET image reconstruction incorporating a median nonlocal means kernel method.

In dynamic positron emission tomography (PET) imaging, the reconstructed image of a single frame often exhibits high noise due to limited counting statistics of projection data. This study proposed a median nonlocal means (MNLM)-based kernel method for dynamic PET image reconstruction. The kernel matrix is derived from median nonlocal means of pre-reconstructed composite images. Then the PET image intensities in all voxels were modeled as a kernel matrix multiplied by coefficients and incorporated into the forward model of PET projection data. Then, the coefficients of each feature were estimated by the maximum likelihood method. Using simulated low-count dynamic data of Zubal head phantom, the quantitative performance of the proposed MNLM kernel method was investigated and compared with the maximum-likelihood method, conventional kernel method with and without median filter, and nonlocal means (NLM) kernel method. Simulation results showed that the MNLM kernel method achieved visual and quantitative accuracy improvements (in terms of the ensemble mean squared error, bias versus variance, and contrast versus noise performances). Especially for frame 2 with the lowest count level of a single frame, the MNLM kernel method achieves lower ensemble mean squared error (10.43%) than the NLM kernel method (13.68%), conventional kernel method with and without median filter (11.88% and 23.50%), and MLEM algorithm (24.77%). The study on real low-dose 18F-FDG rat data also showed that the MNLM kernel method outperformed other methods in visual and quantitative accuracy improvements (in terms of regional noise versus intensity mean performance).

Direct Cellularity Estimation on Breast Cancer Histopathology Images Using Transfer Learning.

Residual cancer burden (RCB) has been proposed to measure the postneoadjuvant breast cancer response. In the workflow of RCB assessment, estimation of cancer cellularity is a critical task, which is conventionally achieved by manually reviewing the hematoxylin and eosin- (H&E-) stained microscopic slides of cancer sections. In this work, we develop an automatic and direct method to estimate cellularity from histopathological image patches using deep feature representation, tree boosting, and support vector machine (SVM), avoiding the segmentation and classification of nuclei. Using a training set of 2394 patches and a test set of 185 patches, the estimations by our method show strong correlation to those by the human pathologists in terms of intraclass correlation (ICC) (0.94 with 95% CI of (0.93, 0.96)), Kendall's tau (0.83 with 95% CI of (0.79, 0.86)), and the prediction probability (0.93 with 95% CI of (0.91, 0.94)), compared to two other methods (ICC of 0.74 with 95% CI of (0.70, 0.77) and 0.83 with 95% CI of (0.79, 0.86)). Our method improves the accuracy and does not rely on annotations of individual nucleus.

[Kinetic cluster and α-divergence-based dynamic myocardial factorial analysis of positron-emission computed tomography images].

OBJECTIVE: We purpose a novel factor analysis method based on kinetic cluster and α-divergence measure for extracting the blood input function and the time-activity curve of the regional tissue from dynamic myocardial positron emission computed tomography(PET) images. METHODS: Dynamic PET images were decomposed into initial factors and factor images by minimizing the α-divergence between the factor model and actual image data. The kinetic clustering as a priori constraint was then incorporated into the model to solve the nonuniqueness problem, and the tissue time-activity curves and the tissue space distributions with physiological significance were generated. RESULTS: The model was applied to the 82RbPET myocardial perfusion simulation data and compared with the traditional model-based least squares measure and the minimal spatial overlap constraint. The experimental results showed that the proposed model performed better than the traditional model in terms of both accuracy and sensitivity. CONCLUSION: This method can select the optimal measure by α value, and incorporate the prior information of the kinetic clustering of PET image pixels to obtain the accurate time-activity curves of the tissue, which has shown good performance in visual evaluation and quantitative evaluation.

Enhanced Performance of Brain Tumor Classification via Tumor Region Augmentation and Partition.

Automatic classification of tissue types of region of interest (ROI) plays an important role in computer-aided diagnosis. In the current study, we focus on the classification of three types of brain tumors (i.e., meningioma, glioma, and pituitary tumor) in T1-weighted contrast-enhanced MRI (CE-MRI) images. Spatial pyramid matching (SPM), which splits the image into increasingly fine rectangular subregions and computes histograms of local features from each subregion, exhibits excellent results for natural scene classification. However, this approach is not applicable for brain tumors, because of the great variations in tumor shape and size. In this paper, we propose a method to enhance the classification performance. First, the augmented tumor region via image dilation is used as the ROI instead of the original tumor region because tumor surrounding tissues can also offer important clues for tumor types. Second, the augmented tumor region is split into increasingly fine ring-form subregions. We evaluate the efficacy of the proposed method on a large dataset with three feature extraction methods, namely, intensity histogram, gray level co-occurrence matrix (GLCM), and bag-of-words (BoW) model. Compared with using tumor region as ROI, using augmented tumor region as ROI improves the accuracies to 82.31% from 71.39%, 84.75% from 78.18%, and 88.19% from 83.54% for intensity histogram, GLCM, and BoW model, respectively. In addition to region augmentation, ring-form partition can further improve the accuracies up to 87.54%, 89.72%, and 91.28%. These experimental results demonstrate that the proposed method is feasible and effective for the classification of brain tumors in T1-weighted CE-MRI.