Deep learning for improving PET/CT attenuation correction by elastic registration of anatomical data.

BACKGROUND: For PET/CT, the CT transmission data are used to correct the PET emission data for attenuation. However, subject motion between the consecutive scans can cause problems for the PET reconstruction. A method to match the CT to the PET would reduce resulting artifacts in the reconstructed images. PURPOSE: This work presents a deep learning technique for inter-modality, elastic registration of PET/CT images for improving PET attenuation correction (AC). The feasibility of the technique is demonstrated for two applications: general whole-body (WB) imaging and cardiac myocardial perfusion imaging (MPI), with a specific focus on respiratory and gross voluntary motion. MATERIALS AND METHODS: A convolutional neural network (CNN) was developed and trained for the registration task, comprising two distinct modules: a feature extractor and a displacement vector field (DVF) regressor. It took as input a non-attenuation-corrected PET/CT image pair and returned the relative DVF between them-it was trained in a supervised fashion using simulated inter-image motion. The 3D motion fields produced by the network were used to resample the CT image volumes, elastically warping them to spatially match the corresponding PET distributions. Performance of the algorithm was evaluated in different independent sets of WB clinical subject data: for recovering deliberate misregistrations imposed in motion-free PET/CT pairs and for improving reconstruction artifacts in cases with actual subject motion. The efficacy of this technique is also demonstrated for improving PET AC in cardiac MPI applications. RESULTS: A single registration network was found to be capable of handling a variety of PET tracers. It demonstrated state-of-the-art performance in the PET/CT registration task and was able to significantly reduce the effects of simulated motion imposed in motion-free, clinical data. Registering the CT to the PET distribution was also found to reduce various types of AC artifacts in the reconstructed PET images of subjects with actual motion. In particular, liver uniformity was improved in the subjects with significant observable respiratory motion. For MPI, the proposed approach yielded advantages for correcting artifacts in myocardial activity quantification and potentially for reducing the rate of the associated diagnostic errors. CONCLUSION: This study demonstrated the feasibility of using deep learning for registering the anatomical image to improve AC in clinical PET/CT reconstruction. Most notably, this improved common respiratory artifacts occurring near the lung/liver border, misalignment artifacts due to gross voluntary motion, and quantification errors in cardiac PET imaging.

Deep learning for whole-body medical image generation.

BACKGROUND: Artificial intelligence (AI) algorithms based on deep convolutional networks have demonstrated remarkable success for image transformation tasks. State-of-the-art results have been achieved by generative adversarial networks (GANs) and training approaches which do not require paired data. Recently, these techniques have been applied in the medical field for cross-domain image translation. PURPOSE: This study investigated deep learning transformation in medical imaging. It was motivated to identify generalizable methods which would satisfy the simultaneous requirements of quality and anatomical accuracy across the entire human body. Specifically, whole-body MR patient data acquired on a PET/MR system were used to generate synthetic CT image volumes. The capacity of these synthetic CT data for use in PET attenuation correction (AC) was evaluated and compared to current MR-based attenuation correction (MR-AC) methods, which typically use multiphase Dixon sequences to segment various tissue types. MATERIALS AND METHODS: This work aimed to investigate the technical performance of a GAN system for general MR-to-CT volumetric transformation and to evaluate the performance of the generated images for PET AC. A dataset comprising matched, same-day PET/MR and PET/CT patient scans was used for validation. RESULTS: A combination of training techniques was used to produce synthetic images which were of high-quality and anatomically accurate. Higher correlation was found between the values of mu maps calculated directly from CT data and those derived from the synthetic CT images than those from the default segmented Dixon approach. Over the entire body, the total amounts of reconstructed PET activities were similar between the two MR-AC methods, but the synthetic CT method yielded higher accuracy for quantifying the tracer uptake in specific regions. CONCLUSION: The findings reported here demonstrate the feasibility of this technique and its potential to improve certain aspects of attenuation correction for PET/MR systems. Moreover, this work may have larger implications for establishing generalized methods for inter-modality, whole-body transformation in medical imaging. Unsupervised deep learning techniques can produce high-quality synthetic images, but additional constraints may be needed to maintain medical integrity in the generated data.

Can 3'-Deoxy-3'-((18)F) Fluorothymidine Out Perform 2-Deoxy-2-((18)F) Fluoro-D-Glucose Positron Emission Tomography/Computed Tomography in the Diagnosis of Cervical Lymphadenopathy in Patients With Oral/Head and Neck Cancer?

PURPOSE: The present study investigated the performance of cellular metabolism imaging with 2-deoxy-2-((18)F) fluoro-D-glucose (FDG) versus cellular proliferation imaging with 3'-deoxy-3'-((18)F) fluorothymidine (FLT) in the detection of cervical lymph node metastases in oral/head and neck cancer. MATERIALS AND METHODS: We conducted a prospective cohort study to assess a head-to-head performance of FLT imaging and clinical FDG imaging for characterizing cervical lymph node metastases in patients with squamous cell carcinoma (SCC) of the oral/head and neck region. The primary predictor variable of the study was the presence of FDG or FLT avidity within the cervical lymph nodes. The primary outcome variable was the histologic presence of metastatic SCC in the cervical lymph nodes. The performance was reported in terms of the sensitivity, specificity, accuracy, and positive and negative predictive values. The overall accuracy for discriminating positive from negative lymph nodes was evaluated as a function of the positron emission tomography (PET) standardized uptake value (SUV). Receiver operating characteristic (ROC) analyses were performed for both tracers. RESULTS: Eleven patients undergoing surgical resection of SCC of the oral/head and neck region underwent preoperative FDG and FLT PET-computed tomography (CT) scans on separate days. The interpretation of the FDG PET-CT imaging resulted in sensitivity, specificity, accuracy, positive predictive value, and negative predictive value of 43.2, 99.5, 94.4, 88.9, and 94.7%, respectively. The sensitivity, specificity, accuracy, positive predictive value, and negative predictive value for FLT PET-CT imaging was 75.7, 99.2, 97.1, 90.3, and 97.7%, respectively. The areas under the curve for the ROC curves were 0.9 and 0.84 for FDG and FLT, respectively. Poor correlation was observed between the SUV for FDG and FLT within the lymph nodes and tumors. CONCLUSION: FLT showed better overall performance for detecting lymphadenopathy on qualitative assessment within the total nodal population. This notwithstanding, FDG SUV performed better for pathologic discrimination within the visible lymph nodes.

Respiratory and cardiac motion correction in positron emission tomography using elastic motion approach for simultaneous abdomen and thorax positron emission tomography-magnetic resonance imaging.

Background: Cardiac and respiratory motions in clinical positron emission tomography (PET) are a major contributor to inaccurate PET quantification and lesion characterisation. In this study, an elastic motion-correction (eMOCO) technique based on mass preservation optical flow is adapted and investigated for positron emission tomography-magnetic resonance imaging (PET-MRI) applications. Methods: The eMOCO technique was investigated in a motion management QA phantom and in twenty-four patients who underwent PET-MRI for dedicated liver imaging and nine patients for cardiac PET-MRI evaluation. Acquired data were reconstructed with eMOCO and gated motion correction techniques at cardiac, respiratory and dual gating modes, and compared to static images. Standardized uptake value (SUV), signal-to-noise ratio (SNR) of lesion activities from each gating mode and correction technique were measured and their means/standard deviation (SD) were compared using 2-ways ANOVA analysis and post-hoc Tukey's test. Results: Lesions' SNR are highly recovered from phantom and patient studies. The SD of the SUV resulted from the eMOCO technique was statistically significantly less (P<0.01) than the SD resulted from conventional gated and static SUVs at the liver, lung and heart. Conclusions: The eMOCO technique was successfully implemented in PET-MRI in a clinical setting and produced the lowest SD compared to gated and static images, and hence provided the least noisy PET images. Therefore, the eMOCO technique can potentially be used on PET-MRI for improved respiratory and cardiac motion correction.

Reproducibility of FDG PET/CT image-based cancer staging and standardized uptake values with simulated reduction of injected FDG dose or acquisition time.

18F-fluorodeoxyglucose (FDG) PET/CT is widely used for oncologic imaging. This study aimed to evaluate, using data simulation, if reduction of injected FDG dose or PET acquisition time could be technically feasible when utilizing a sensitive commercial PET/CT imaging system, without sacrificing image quality, image-based staging accuracy, or standardized uptake value (SUV) accuracy. De-identified, standard of care oncologic FDG PET/CT datasets from 83 adults with lymphoma, lung carcinoma or breast carcinoma were retrospectively analyzed. All images had been acquired using clinical standard dose and acquisition time on a single PET/CT system. The list mode datasets were retrospectively software reprocessed to achieve undersampling of counts, thus simulating the effect of shorter PET acquisition time or lower injected FDG dose. The simulated reduced-count images were reviewed and compared with full-count images to assess and compare qualitative (subjective image quality, stage stability) and semi-quantitative (image noise, SUVmax stability, signal-to-noise and contrast-to-noise ratios within index lesions driving cancer stage) parameters. While simulated reduced-count images had measurably greater noise, there appeared to be no significant loss of image-based staging accuracy nor SUVmax reproducibility down to simulated FDG dose of 0.05 mCi/kg at continuous bed motion rate of 1.1 mm/sec. This retrospective simulation study suggests that a modest reduction of either injected FDG dose or emission scan time might be feasible in this limited oncologic population scanned on a single PET/CT system. Verification of these results with prospectively acquired images using actual low injected FDG activity and/or short imaging time is recommended.

Data-driven respiratory gating based on localized diaphragm sensing in TOF PET.

It is important to measure the respiratory cycle in positron emission tomography (PET) to enhance the contrast of the tumor as well as the accuracy of its localization in organs such as the lung and liver. Several types of data-driven respiratory gating methods, such as center of mass and principal component analysis, have been developed to directly measure the breathing cycle from PET images and listmode data. However, the breathing cycle is still hard to detect in low signal-to-noise ratio (SNR) data, particularly in low dose PET/CT scans. To address this issue, a time-of-flight (TOF) PET is currently utilized for the data-driven respiratory gating because of its higher SNR and better localization of the region of interest. To further improve the accuracy of respiratory gating with TOF information, we propose an accurate data-driven respiratory gating method, which retrospectively derives the respiratory signal using a localized sensing method based on a diaphragm mask in TOF PET data. To assess the accuracy of the proposed method, the performance is evaluated with three patient datasets, and a pressure-belt signal as the ground truth is compared. In our experiments, we validate that the respiratory signal using the proposed data-driven gating method is well matched to the pressure-belt respiratory signal with less than 5% peak time errors and over 80% trace correlations. Based on gated signals, the respiratory-gated image of the proposed method provides more clear edges of organs compared to images using conventional non-TOF methods. Therefore, we demonstrate that the proposed method can achieve improvements for the accuracy of gating signals and image quality.

Supervised learning with cyclegan for low-dose FDG PET image denoising.

PET imaging involves radiotracer injections, raising concerns about the risk of radiation exposure. To minimize the potential risk, one way is to reduce the injected tracer. However, this will lead to poor image quality with conventional image reconstruction and processing. In this paper, we proposed a supervised deep learning model, CycleWGANs, to boost low-dose PET image quality. Validations were performed on a low dose dataset simulated from a real dataset with biopsy-proven primary lung cancer or suspicious radiological abnormalities. Low dose PET images were reconstructed on reduced PET raw data by randomly discarding events in the PET list mode data towards the count level of 1 million. Traditional image denoising methods (Non-Local Mean (NLM) and block-matching 3D(BM3D)) and two recently-published deep learning methods (RED-CNN and 3D-cGAN) were included for comparisons. At the count level of 1 million (true counts), the proposed model can accurately estimate full-dose PET image from low-dose input image, which is superior to the other four methods in terms of the mean and maximum standardized uptake value (SUVmean and SUVmax) bias for lesions and normal tissues. The bias of SUV (SUVmean, SUVmax) for lesions and normal tissues are (-2.06±3.50%,-0.84±6.94%) and (-0.45±5.59%, N/A) in the estimated PET images, respectively. However, the RED-CNN achieved the best score in traditional metrics, such as structure similarity (SSIM), peak signal to noise ratio (PSNR) and normalized root mean square error (NRMSE). Correlation and profile analyses have successfully explained this phenomenon and further suggested that our method could effectively preserve edge and also SUV values than RED-CNN, 3D-cGAN and NLM with a slightly higher noise.

Capsinoids activate brown adipose tissue (BAT) with increased energy expenditure associated with subthreshold 18-fluorine fluorodeoxyglucose uptake in BAT-positive humans confirmed by positron emission tomography scan.

Background: Capsinoids are reported to increase energy expenditure (EE) via brown adipose tissue (BAT) stimulation. However, imaging of BAT activation by capsinoids remains limited. Because BAT activation is a potential therapeutic strategy for obesity and related metabolic disorders, we sought to prove that capsinoid-induced BAT activation can be visualized by 18-fluorine fluorodeoxyglucose (18F-FDG) positron emission tomography (PET). Objective: We compared capsinoids and cold exposure on BAT activation and whole-body EE. Design: Twenty healthy participants (8 men, 12 women) with a mean age of 26 y (range: 21-35 y) and a body mass index (kg/m2) of 21.7 (range: 18.5-26.0) underwent 18F-FDG PET and whole-body calorimetry after ingestion of 12 mg capsinoids or ≤2 h of cold exposure (∼14.5°C) in a crossover design. Mean standardized uptake values (SUVs) of the region of interest and BAT volumes were calculated. Blood metabolites were measured before and 2 h after each treatment. Results: All of the participants showed negligible 18F-FDG uptake post-capsinoid ingestion. Upon cold exposure, 12 participants showed avid 18F-FDG uptake into supraclavicular and lateral neck adipose tissues (BAT-positive group), whereas the remaining 8 participants (BAT-negative group) showed undetectable uptake. Capsinoids and cold exposure increased EE, although cold induced a 2-fold increase in whole-body EE and higher fat oxidation, insulin sensitivity, and HDL cholesterol compared with capsinoids. Conclusions: Capsinoids only increased EE in BAT-positive participants, which suggests that BAT mediates EE evoked by capsinoids. This implies that capsinoids stimulate BAT to a lesser degree than cold exposure as evidenced by 18F-FDG uptake below the presently accepted SUV thresholds defining BAT activation. This trial was registered at www.clinicaltrials.gov as NCT02964442.

Quantitative Accuracy and Lesion Detectability of Low-Dose 18F-FDG PET for Lung Cancer Screening.

Lung cancer remains responsible for more deaths worldwide than any other cancer, but recently there has been a significant shift in the clinical paradigm regarding the initial management of subjects at high risk for this disease. Low-dose CT has demonstrated significant improvements over planar x-ray screening for patient prognoses and is now performed in the United States. Specificity of this modality, however, is poor, and the additional information from PET has the potential to improve its accuracy. Routine screening requires consideration of the effective dose delivered to the patient, and this work investigates image quality of PET for low-dose conditions, in the context of lung lesion detectability. Reduced radiotracer doses were simulated by randomly discarding counts from clinical lung cancer scans acquired in list-mode. Bias and reproducibility of lesion activity values were relatively stable even at low total counts of around 5 million trues. Additionally, numeric observer models were developed and trained with the results of 2 physicians and 3 postdoctoral researchers with PET experience in a detection task; detection sensitivity of the observers was well correlated with lesion signal-to-noise ratio. The models were used prospectively to survey detectability of lung cancer lesions, and the findings suggested a lower limit around 10 million true counts for maximizing performance. Under the acquisition parameters used in this study, this translates to an effective patient dose of less than 0.4 mSv, potentially allowing a complete low-dose PET/CT lung screening scan to be obtained under 1 mSv.

Quantitative 18F-fluorocholine positron emission tomography for prostate cancer: correlation between kinetic parameters and Gleason scoring.

BACKGROUND: The use of radiolabeled choline as a positron emission tomography (PET) agent for imaging primary tumors in the prostate has been evaluated extensively over the past two decades. There are, however, conflicting reports of its sensitivity and the relationship between choline PET imaging and disease staging is not fully understood. Moreover, relatively few studies have investigated the correlation between tracer uptake and histological tumor grade. This work quantified 18F-fluorocholine in tumor and healthy prostate tissue using pharmacokinetic modeling and stratified uptake parameters by histology grade. Additionally, the effect of scan time on the estimation of the kinetic exchange rate constants was evaluated, and the tracer influx parameters from full compartmental analysis were compared to uptake values quantified by Patlak and standardized uptake value (SUV) analyses. 18F-fluorocholine was administered as a 222 MBq bolus injection to ten patients with biopsy-confirmed prostate tumors, and dynamic PET data were acquired for 60 min. Image-derived arterial input functions were scaled by discrete blood samples, and a 2-tissue, 4-parameter model accounting for blood volume (2T4k+Vb) was used to perform fully quantitative compartmental modeling on tumor, healthy prostate, and muscle tissue. Subsequently, all patients underwent radical prostatectomy, and histological analyses were performed on the prostate specimens; kinetic parameters for tumors were stratified by Gleason score. Correlations were investigated between compartmental K 1 and K i parameters and SUV and Patlak slope; the effect of scan time on parameter bias was also evaluated. RESULTS: Choline activity curves in seven tumors, eight healthy prostate regions, and nine muscle regions were analyzed. Net tracer influx was generally higher in tumor relative to healthy prostate, with the values in the highest grade tumors markedly higher than those in lower grade tumors. Influx terms from Patlak and full compartmental modeling showed good correlation within individual tissue groups. Kinetic parameters calculated from the entire 60-min scan data were accurately reproduced from the first 30 min of acquired data (R 2 ≈ 0.9). CONCLUSIONS: Strong correlations were observed between K i and Patlak slope in tumor tissue, and K 1 and SUV were also correlated but to a lesser degree. Reliable estimates of all kinetic parameters can be achieved from the first 30 min of dynamic 18F-choline data. Although SUV, K 1, K i, and Patlak slope were found to be poor differentiators of low-grade tumor compared to healthy prostate tissue, they are strong indicators of aggressive disease.

Initial assessment of image quality for low-dose PET: evaluation of lesion detectability.

In the context of investigating the potential of low-dose PET imaging for screening applications, we developed methods to assess small lesion detectability as a function of the number of counts in the scan. We present here our methods and preliminary validation using tuberculosis cases. FDG-PET data from seventeen patients presenting diffuse hyper-metabolic lung lesions were selected for the study, to include a wide range of lesion sizes and contrasts. Reduced doses were simulated by randomly discarding events in the PET list mode, and ten realizations at each simulated dose were generated and reconstructed. The data were grouped into 9 categories determined by the number of included true events, from >40 M to <250 k counts. The images reconstructed from the original full statistical set were used to identify lung lesions, and each was, at every simulated dose, quantified by 6 parameters: lesion metabolic volume, lesion-to-background contrast, mean lesion tracer uptake, standard deviation of activity measurements (across realizations), lesion signal-to-noise ratio (SNR), and Hotelling observer SNR. Additionally, a lesion-detection task including 550 images was presented to several experienced image readers for qualitative assessment. Human observer performances were ranked using receiver operating characteristic analysis. The observer results were correlated with the lesion image measurements and used to train mathematical observer models. Absolute sensitivities and specificities of the human observers, as well as the area under the ROC curve, showed clustering and performance similarities among images produced from 5 million or greater counts. The results presented here are from a clinically realistic but highly constrained experiment, and more work is needed to validate these findings with a larger patient population.

Effect of time-of-flight and point spread function modeling on detectability of myocardial defects in PET.

PURPOSE: A study was designed to investigate the impact of time-of-flight (TOF) and point spread function (PSF) modeling on the detectability of myocardial defects. METHODS: Clinical FDG-PET data were used to generate populations of defect-present and defect-absent images. Defects were incorporated at three contrast levels, and images were reconstructed by ordered subset expectation maximization (OSEM) iterative methods including ordinary Poisson, alone and with PSF, TOF, and PSF+TOF. Channelized Hotelling observer signal-to-noise ratio (SNR) was the surrogate for human observer performance. RESULTS: For three iterations, 12 subsets, and no postreconstruction smoothing, TOF improved overall defect detection SNR by 8.6% as compared to its non-TOF counterpart for all the defect contrasts. Due to the slow convergence of PSF reconstruction, PSF yielded 4.4% less SNR than non-PSF. For reconstruction parameters (iteration number and postreconstruction smoothing kernel size) optimizing observer SNR, PSF showed larger improvement for faint defects. The combination of TOF and PSF improved mean detection SNR as compared to non-TOF and non-PSF counterparts by 3.0% and 3.2%, respectively. CONCLUSIONS: For typical reconstruction protocol used in clinical practice, i.e., less than five iterations, TOF improved defect detectability. In contrast, PSF generally yielded less detectability. For large number of iterations, TOF+PSF yields the best observer performance.

Complementary frame reconstruction: a low-biased dynamic PET technique for low count density data in projection space.

A new data handling method is presented for improving the image noise distribution and reducing bias when reconstructing very short frames from low count dynamic PET acquisition. The new method termed 'Complementary Frame Reconstruction' (CFR) involves the indirect formation of a count-limited emission image in a short frame through subtraction of two frames with longer acquisition time, where the short time frame data is excluded from the second long frame data before the reconstruction. This approach can be regarded as an alternative to the AML algorithm recently proposed by Nuyts et al, as a method to reduce the bias for the maximum likelihood expectation maximization (MLEM) reconstruction of count limited data. CFR uses long scan emission data to stabilize the reconstruction and avoids modification of algorithms such as MLEM. The subtraction between two long frame images, naturally allows negative voxel values and significantly reduces bias introduced in the final image. Simulations based on phantom and clinical data were used to evaluate the accuracy of the reconstructed images to represent the true activity distribution. Applicability to determine the arterial input function in human and small animal studies is also explored. In situations with limited count rate, e.g. pediatric applications, gated abdominal, cardiac studies, etc., or when using limited doses of short-lived isotopes such as 15O-water, the proposed method will likely be preferred over independent frame reconstruction to address bias and noise issues.

Myocardial defect detection using PET-CT: phantom studies.

It is expected that both noise and activity distribution can have impact on the detectability of a myocardial defect in a cardiac PET study. In this work, we performed phantom studies to investigate the detectability of a defect in the myocardium for different noise levels and activity distributions. We evaluated the performance of three reconstruction schemes: Filtered Back-Projection (FBP), Ordinary Poisson Ordered Subset Expectation Maximization (OP-OSEM), and Point Spread Function corrected OSEM (PSF-OSEM). We used the Channelized Hotelling Observer (CHO) for the task of myocardial defect detection. We found that the detectability of a myocardial defect is almost entirely dependent on the noise level and the contrast between the defect and its surroundings.

Clinical impact of time-of-flight and point response modeling in PET reconstructions: a lesion detection study.

Time-of-flight (TOF) and point spread function (PSF) modeling have been shown to improve PET reconstructions, but the impact on physicians in the clinical setting has not been thoroughly investigated. A lesion detection and localization study was performed using simulated lesions in real patient images. Four reconstruction schemes were considered: ordinary Poisson OSEM (OP) alone and combined with TOF, PSF, and TOF + PSF. The images were presented to physicians experienced in reading PET images, and the performance of each was quantified using localization receiver operating characteristic. Numerical observers (non-prewhitening and Hotelling) were used to identify optimal reconstruction parameters, and observer SNR was compared to the performance of the physicians. The numerical models showed good agreement with human performance, and best performance was achieved by both when using TOF + PSF. These findings suggest a large potential benefit of TOF + PSF for oncology PET studies, especially in the detection of small, low-intensity, focal disease in larger patients.