Automatic Mismatch Correction and Motion Compensation for Free-breathing PET/CT

ObjectiveWe aimed to develop a standardized method to investigate the relationship between estimated brain age and regional morphometric features, meeting the criteria for simplicity, generalization, and intuitive interpretability.MethodsWe utilized T1-weighted magnetic resonance imaging (MRI) data from the Cambridge Centre for Ageing and Neuroscience project (N = 609) and employed a support vector regression method to train a brain age model. The pre-trained brain age model was applied to the dataset of the brain development project (N = 547). Kraskov (KSG) estimator was used to compute the mutual information (MI) value between brain age and regional morphometric features, including gray matter volume (GMV), white matter volume (WMV), cerebrospinal fluid (CSF) volume, and cortical thickness (CT).ResultsAmong four types of brain features, GMV had the highest MI value (8.71), peaking in the pre-central gyrus (0.69). CSF volume was ranked second (7.76), with the highest MI value in the cingulate (0.87). CT was ranked third (6.22), with the highest MI value in superior temporal gyrus (0.53). WMV had the lowest MI value (4.59), with the insula showing the highest MI value (0.53). For brain parenchyma, the volume of the superior frontal gyrus exhibited the highest MI value (0.80).ConclusionThis is the first demonstration that MI value between estimated brain age and morphometric features may serve as a benchmark for assessing the regional contributions to estimated brain age. Our findings highlighted that both GMV and CSF are the key features that determined the estimated brain age, which may add value to existing computational models of brain age.Critical relevance statementMutual information (MI) analysis reveals gray matter volume (GMV) and cerebrospinal fluid (CSF) volume as pivotal in computing individuals’ brain age.Key PointsMutual information (MI) interprets estimated brain age with morphometric features.Gray matter volume in the pre-central gyrus has the highest MI value for estimated brain age.Cerebrospinal fluid volume in the cingulate has the highest MI value.Regarding brain parenchymal volume, the superior frontal gyrus has the highest MI value.The value of mutual information underscores the key brain regions related to brain age.Graphical

The PET/MRI hybrid system became eligible for staging the diagnosis of malignant tumors, metastasis, and recurrence using 18F-Fluorodeoxyglucose (FDG). Inserting PET in an MRI is technically challenging and requires PET detectors to be compatible with the magnetic field. The key technology that enables this is the semiconductor-based photomultiplier tube. The advantage of the PET/MRI hybrid system is not simply that MRI has replaced computed tomography (CT), but rather that PET and MRI data can be acquired simultaneously. The high-precision fusion images show improved diagnostic accuracy. Positional misalignment due to movements, such as respiratory motion, as seen on PET/CT, can be minimized through simultaneous acquisition, and higher-precision fusion images can be obtained by performing PET and MRI acquisitions under respiratory gating. One of the disadvantages of PET/MRI over PET/CT is that CT cannot be used for gamma-ray attenuation correction, which is necessary for PET image reconstruction. Although a method has been established to generate a pseudo-CT image from MRI for attenuation correction, generating a pseudo-CT image that includes bone components is particularly difficult. In PET/MRI systems approved for clinical use, bone components are not considered for attenuation correction outside the head region. Recently, active research has been conducted utilizing deep learning to generate pseudo-CT images with bone components from MRI, regardless of the anatomical region, or to perform attenuation correction using PET data alone. Recently, new PET tracers have been approved in Japan. As PET/MRI can produce highly accurate fusion images with MRI, these tracers and target lesions are likely to benefit more from PET/MRI than from PET/CT. As such, the scope of PET/MRI applications is expected to continue to expand, and efforts should be made to promote its widespread adoption so that more facilities can utilize this technology.

CT images acquired during deep-inspiration breath-holding (DIBH) have higher image quality than free-breathing (FB) CT images as they are free from respiratory motion artifacts. However, there exists significant mismatch between DIBH CT images and FB PET images since the displacement of organs in the thorax and abdomen at DIBH is significantly larger than that during regular FB. Our goal is to correct for the large mismatch between FB PET and DIBH CT in PET/CT imaging. The list-mode PET data were divided into four equal-count frames using amplitude based data-driven respiratory gating technique. Frame 1 and Frame 4 corresponded to end-expiration and end-inspiration respectively, and they were reconstructed using OSEM image reconstruction algorithm. To avoid attenuation-activity mismatch artifacts, the attenuation map used in attenuation correction was modified by filling the lung region with attenuation coefficient of soft tissue. 3D motion vector field from Frame 4 to Frame 1 was calculated from the reconstructed PET images using B-spline image registration algorithm. Assuming the respiratory motion during FB is consistent during FB and DIBH, the motion vector field from DIBH to end-inspiration phase is modeled by scaling the calculated respiratory motion from Frame 4 to Frame 1. The scalar was determined by maximizing the mutual information between transformed DIBH attenuation map and PET image at Frame 4. DIBH attenuation map was transformed using the scaled motion vector field from Frame 4 to Frame 1 to obtain attenuation map at Frame 4. Attenuation maps for other frames were obtained by transforming attenuation map at Frame 4 using corresponding motion vector fields. The method was applied to clinical patient data acquired using DIBH CT and FB PET protocol. While DIBH attenuation map creates significant mismatch artifacts in PET images, the proposed method was able to generate attenuation maps matching FB PET frames and reduce mismatch artifacts.

A critical question in the development of combined MR/PET scanners is whether MR can provide the tissue attenuation data required for PET reconstruction. Unfortunately, MR images are often unable to encompass the entire patient. The resulting truncation in the transverse plane leads to incomplete attenuation maps, causing artifacts in the reconstructed PET image. This article describes the experiments performed to quantify these artifacts. A method to compensate the missing data was evaluated to determine whether software correction is possible or whether additional transmission hardware has to be included in the scanner. Three studies were made. First, simulated PET data were used to quantify the bias due to an incomplete attenuation map. A set of spherical lesions was simulated in the lungs and mediastinum of a patient. The data were reconstructed with complete and partial attenuation maps and the uptake differences were evaluated. Second, clinical data from PET/CT oncology patients were used. To reproduce the expected conditions in an MR/PET scanner, only patients scanned with the arms resting along the body were considered. These scans were then used to create maps of the reconstruction bias due to field of view (FOV) limitations. Lastly, a PET reconstruction with incomplete attenuation data was evaluated as a means to obtain attenuation information beyond the MR FOV. The patient outline was automatically segmented with a three-dimensional snake algorithm and used to fill the truncated data in the attenuation map. Average bias up to 15% and local biases up to 50% were estimated when PET data were reconstructed with incomplete attenuation information. Completing the attenuation map with data extracted from a PET prereconstruction globally reduced these biases to below 10%. This correction proved to be tolerant to inaccuracies in positioning and attenuation values. However, local artifacts up to 20% could still be found near the edges of the MR FOV. MR FOV restrictions can indeed make the reconstructed PET data unacceptable for diagnostic purposes. Biases can be globally compensated by automatic preprocessing of the attenuation map. However, inaccuracies in the correction will result in small artifacts near the periphery of the image that could lead to false-positive findings.

A multi-model fusion strategy for multivariate calibration using near and mid-infrared spectra of samples from brewing industry

Non Double-Hit Diffuse Large B Cell Lymphoma Treated with R-CHOP Has Excellent Overall Survival If Interim Scan Shows Partial or Complete Response

Objectives: To develop a machine learning-based model to select the most important features from a high-dimensional dataset to classify the patterns at high accuracy and reduce their dimensionality. Methods: The proposed feature selection method (FSIFC) forms and combines feature clusters incrementally and produces feature subsets each time. The method uses K-means clustering and Mutual Information (MI) to refine the feature selection process iteratively. Initially, two clusters of features are formed using K-means clustering (K=2) by taking features as the basis of clustering instead of taking the patterns (a traditional way). From these two clusters, the features with the highest MI value in each cluster are kept in a feature subset. Classification accuracies (CA) of the feature subset are calculated using three classifiers namely Support Vector Machines (SVM), Random Forest (RF), and k-nearest Neighbor (knn). The process is repeated by incrementing the value of K i.e. number of clusters; until a maximum user-defined value of K is reached. The best value of CA obtained from these trials is recorded and the corresponding feature set is finally accepted. Findings: The proposed method is demonstrated using ten datasets and the results are compared with the existing published results using three classifiers to determine the method's performance. The ten datasets are classified with average CAs of 92.72%, 93.13%, and 91.5%, using the SVM, RF, and K-NN classifiers respectively. The proposed method selects a maximum of thirty features from the datasets. In terms of selecting the most effective and the smallest feature sets, the proposed method outperforms eight other feature selection methods considering CAs. Novelty: The proposed model applies feature reduction using combined feature clustering and filter methods in an incremental way. This provides an improved selection of relevant features while removing those which are irrelevant at different trials. Keywords: Feature selection, High-dimensional datasets, K-means algorithm, Mutual information, Machine learning

Spatial normalization of multiple sclerosis brain MRI data depends on analysis method and software package

In dual-modality PET/CT systems, the CT scan provides the attenuation map for PET attenuation correction. The current clinical practice of obtaining a single helical CT scan provides only a snapshot of the respiratory cycle, whereas PET occurs over multiple respiratory cycles. Misalignment of the attenuation map and emission image because of respiratory motion causes errors in the attenuation correction factors and artifacts in the attenuation-corrected PET image. To rectify this problem, we evaluated the use of cine CT, which acquires multiple low-dose CT images during a respiratory cycle. We evaluated the average and the intensity-maximum image of cine CT for cardiac PET attenuation correction. Cine CT data and cardiac PET data were acquired from a cardiac phantom and from multiple patient studies. The conventional helical CT, cine CT, and PET data of an axially translating phantom were evaluated with and without respiratory motion. For the patient studies, we acquired 2 cine CT studies for each PET acquisition in a rest-stress (13)N-ammonia protocol. Three readers visually evaluated the alignment of 74 attenuation image sets versus the corresponding emission image and determined whether the alignment provided acceptable or unacceptable attenuation-corrected PET images. In the phantom study, the attenuation correction from helical CT caused a major artifactual defect in the lateral wall on the PET image. The attenuation correction from the average and from the intensity-maximum cine CT images reduced the defect by 20% and 60%, respectively. In the patient studies, 77% of the cases using the average of the cine CT images had acceptable alignment and 88% of the cases using the intensity maximum of the cine CT images had acceptable alignment. Cine CT offers an alternative to helical CT for compensating for respiratory motion in the attenuation correction of cardiac PET studies. Phantom studies suggest that the average and the intensity maximum of the cine CT images can reduce potential respiration-induced misalignment errors in attenuation correction. Patient studies reveal that cine CT provides acceptable alignment in most cases and suggest that the intensity-maximum cine image offers a more robust alternative to the average cine image.

Joint reconstruction of activity and attenuation maps from emission data only, while being a long-standing problem in emission tomography, has been gaining recent interests because of its application to MR-based attenuation correction in PET/MR scanners where CT images are not available. Furthermore, recent studies showed that TOF (time-of-flight) information can substantially reduce, or completely remove in theory, crosstalk artifacts, which had been one of the hurdles preventing joint reconstruction techniques from being used clinically. Nonetheless, estimating both activity and attenuation from TOF emission data is a computationally challenging nonconvex optimization problem with high-dimensional data size. Therefore, as a tool for investigating and optimizing the joint estimation techniques for clinical use, we need numerical algorithms that are derived in a principled way and guaranteed to converge to a solution. Here, in a PL (penalized-likelihood) framework, we present a block alternating MM (minorization maximization) algorithm, which is provably globally convergent although the PL objective function is nonconcave. By using linear parameterization of attenuation maps, the algorithm applies to a variety of scenarios, depending on the type of and the degree of prior information, in a unified way. In addition, we provide a proof of the uniqueness of solutions to joint estimation problems for a continuous-space TOF PET system whose TOF kernels do not need to be Gaussian.

Purpose: We propose to characterize the spatial distribution of PET‐derived voxel statistics within anatomically‐defined Regions Of Interest (ROIs) as a means to characterize disease‐related changes. Methods: We used 3D moment invariants (3DMIs) to characterize the spatial distribution of PET data ([11C]Raclopride, [11C]Tetrabenazine and [18F]Fluorodopa) recorded from subjects with Parkinsonˈs Disease (PD) and healthy controls. 3DMIs are mathematical spatial descriptors designed to be invariant to scaling, translation and rotation. In fMRI studies, assessing the spatial characteristics of voxel‐based statistics has recently been shown to utilized be a powerful and sensitive method to characterize brain activation. Crucially, this allows characterization of the spatial characteristics of activation without the need to warp brain images to a common brain template. Analogously, we propose characterizing the spatial distribution of PET‐derived voxel statistics within anatomically‐defined ROIs using 3DMIs. The 3DMIs used here were a combination of terms describing spatial variance, skewness and kurtosis. Subjects also underwent MRI scans so that the T1‐weighted MRI data could be used to anatomically delineate basal ganglia ROIs to act as binary masks on the PET data. Results: 3DMIs were found to accurately describe the “3D texture” of PET images despite changes in the size and orientation of these regions across subjects. In addition, we were able to find differences in the 3DMIs of PD patients distinct from those of healthy volunteers. These changes suggest that disease‐related variations in the spatial distribution measured using PET can be quantitatively described with the proposed method Conclusions: This method shows great promise to extract additional information from PET data with a wealth of potential applications to disease diagnosis, staging, treatment assessment and more. The quantification of the observed disease‐related changes for PD subjects is currently under way.

Dual modality PET-CT imaging provides aligned anatomical (CT) and functional (PET) images in a single scanning session, which can potentially be used to improve image segmentation of PET-CT data. The ability to distinguish structures for segmentation is a function of structure and modality and varies across voxels. Thus optimal contribution of a particular modality to segmentation is spatially variant. Existing segmentation algorithms, however, seldom account for this characteristic of PET-CT data and the results using these algorithms are not optimal. In this study, we propose a relative discrimination index (RDI) to characterize the relative abilities of PET and CT to correctly classify each voxel into the correct structure for segmentation. The definition of RDI is based on the information entropy of the probability distribution of the voxel's class label. If the class label derived from CT data for a particular voxel has more certainty than that derived from PET data, the corresponding RDI will have a higher value. We applied the RDI matrix to balance adaptively the contributions of PET and CT data to segmentation of brain PET-CT images on a voxel-by-voxel basis, with the aim to give the modality with higher discriminatory power a larger weight. The resultant segmentation approach is distinguished from traditional approaches by its innovative and adaptive use of the dual-modality information. We compared our approach to the non-RDI version and two commonly used PET-only based segmentation algorithms for simulation and clinical data. Our results show that the RDI matrix markedly improved PET-CT image segmentation.

Misregistration between CT and PET data can result in mis-localization and inaccurate quantification of functional uptake in whole body PET/CT imaging. This problem is exacerbated when an abnormal inspiration occurs during the free-breathing helical CT (FB CT) used for attenuation correction of PET data. In data-driven gated (DDG) PET, the data selected for reconstruction is typically derived from the end-expiration (EE) phase of the breathing cycle, making this potential issue worse. The objective of this study is to develop a deformable image registration (DIR)-based respiratory motion model to improve the registration and quantification between misregistered FB CT and PET. Twenty-two whole-body 18 F-FDG PET/CT scans encompassing 48 lesions in misregistered regions were analyzed in this study. End-inspiration (EI) and EE PET data were derived from -10% to 15% and 30% to 80% of the breathing cycle, respectively. DIR was used to estimate a motion model from the EE to EI phase of the PET data. The model was then used to generate PET images at any phase of up to four times the amplitude of motion between EE and EI for correlation with the misregistered FB CT. Once a matched phase of the FB CT was determined, FB CT was deformed to a pseudo CT at the EE phase (DIR CT). DIR CT was compared with the ground truth DDG CT for AC and localization of the DDG PET. Between DDG PET/FB CT and DDG PET/DIR CT, a significant increase in ∆%SUV was observed (p<0.01), with median values elevating from 26.7% to 42.4%. This new method was most effective for lesions ≤3cm proximal to the diaphragm (p<0.001) but showed decreasing efficacy as the distance increased. When FB CT was severely misregistered with DDG PET (>3cm), DDG PET/DIR CT outperformed DDG PET/FB CT alone (p<0.05). Even when patients showed varied breathing patterns during the PET/CT scan, DDG PET/DIR CT still surpassed the efficiency of DDG PET/FB CT (p<0.01). Though DDG PET/DIR CT couldn't match the performance of the DDG PET/CT ground truth (42.4%vs. 53.6%, p<0.01), it reached 84% of its quantification, demonstrating good agreement and a strong overall correlation (regression coefficient of 0.94, p<0.0001). In some cases, anatomical distortion and blurring, and misregistration error were observed in DIR CT, rendering it still unable to correct inaccurate localization near the boundaries of two organs. Based on the motion model derived from gated PET data, DIR CT can significantly improve the quantification and localization of DDG PET. This approach can achieve a performance level of about 84% of the ground truth established by DDG PET/CT. These results show that self-gated PET and DIR CT may offer an alternative clinical solution to DDG PET and FB CT for quantification without the need for additional cine-CT imaging. DIR CT was at times inferior to DDG CT due to some distortion and blurring of anatomy and misregistration error.

Although positron emission tomography-computed tomography (PET-CT) images have been widely used, it is still challenging to accurately segment the lung tumor. The respiration, movement and imaging modality lead to large modality discrepancy of the lung tumors between PET images and CT images. To overcome these difficulties, a novel network is designed to simultaneously obtain the corresponding lung tumors of PET images and CT images. The proposed network can fuse the complementary information and preserve modality-specific features of PET images and CT images. Due to the complementarity between PET images and CT images, the two modality images should be fused for automatic lung tumor segmentation. Therefore, cross modality decoding blocks are designed to extract modality-specific features of PET images and CT images with the constraints of the other modality. The edge consistency loss is also designed to solve the problem of blurred boundaries of PET images and CT images. The proposed method is tested on 126 PET-CT images with non-small cell lung cancer, and Dice similarity coefficient scores of lung tumor segmentation reach 75.66 ± 19.42 in CT images and 79.85 ± 16.76 in PET images, respectively. Extensive comparisons with state-of-the-art lung tumor segmentation methods have also been performed to demonstrate the superiority of the proposed network.

The impact of a method for MR-based respiratory motion correction of PET data on lesion visibility and quantification in patients with oncologic findings in the lung was evaluated. Twenty patients with one or more lesions in the lung were included. Hybrid imaging was performed on an integrated PET/MR system using 18F-FDG as radiotracer. The standard thoracic imaging protocol was extended by a free-breathing self-gated acquisition of MR data for motion modelling. PET data was acquired simultaneously in list-mode for 5-10 mins. One experienced radiologist and one experienced nuclear medicine specialist evaluated and compared the post-processed data in consensus regarding lesion visibility (scores 1–4, 4 being best), image noise levels (scores 1–3, 3 being lowest noise), SUVmean and SUVmax. Motion-corrected (MoCo) images were additionally compared with gated images. Non-motion-corrected free-breathing data served as standard of reference in this study. Motion correction generally improved lesion visibility (3.19 ± 0.63) and noise ratings (2.95 ± 0.22) compared to uncorrected (2.81 ± 0.66 and 2.95 ± 0.22, respectively) or gated PET data (2.47 ± 0.93 and 1.30 ± 0.47, respectively). Furthermore, SUVs (mean and max) were compared for all methods to estimate their respective impact on the quantification. Deviations of SUVmax were smallest between the uncorrected and the MoCo lesion data (average increase of 9.1% of MoCo SUVs), while SUVmean agreed best for gated and MoCo reconstructions (MoCo SUVs increased by 1.2%). The studied method for MR-based respiratory motion correction of PET data combines increased lesion sharpness and improved lesion activity quantification with high signal-to-noise ratio in a clinical setting. In particular, the detection of small lesions in moving organs such as the lung and liver may thus be facilitated. These advantages justify the extension of the PET/MR imaging protocol by 5–10 minutes for motion correction.