Search

Purpose: To develop a two-stage three-dimensional (3D) convolutional neural networks (CNNs) for fully automated volumetric segmentation of pancreas on computed tomography (CT) and to further evaluate its performance in the context of intra-reader and inter-reader reliability at full dose and reduced radiation dose CTs on a public dataset., Methods: A dataset of 1994 abdomen CT scans (portal venous phase, slice thickness ≤ 3.75-mm, multiple CT vendors) was curated by two radiologists (R1 and R2) to exclude cases with pancreatic pathology, suboptimal image quality, and image artifacts (n = 77). Remaining 1917 CTs were equally allocated between R1 and R2 for volumetric pancreas segmentation [ground truth (GT)]. This internal dataset was randomly divided into training (n = 1380), validation (n = 248), and test (n = 289) sets for the development of a two-stage 3D CNN model based on a modified U-net architecture for automated volumetric pancreas segmentation. Model's performance for pancreas segmentation and the differences in model-predicted pancreatic volumes vs GT volumes were compared on the test set. Subsequently, an external dataset from The Cancer Imaging Archive (TCIA) that had CT scans acquired at standard radiation dose and same scans reconstructed at a simulated 25% radiation dose was curated (n = 41). Volumetric pancreas segmentation was done on this TCIA dataset by R1 and R2 independently on the full dose and then at the reduced radiation dose CT images. Intra-reader and inter-reader reliability, model's segmentation performance, and reliability between model-predicted pancreatic volumes at full vs reduced dose were measured. Finally, model's performance was tested on the benchmarking National Institute of Health (NIH)-Pancreas CT (PCT) dataset., Results: Three-dimensional CNN had mean (SD) Dice similarity coefficient (DSC): 0.91 (0.03) and average Hausdorff distance of 0.15 (0.09) mm on the test set. Model's performance was equivalent between males and females (P = 0.08) and across different CT slice thicknesses (P > 0.05) based on noninferiority statistical testing. There was no difference in model-predicted and GT pancreatic volumes [mean predicted volume 99 cc (31cc); GT volume 101 cc (33 cc), P = 0.33]. Mean pancreatic volume difference was -2.7 cc (percent difference: -2.4% of GT volume) with excellent correlation between model-predicted and GT volumes [concordance correlation coefficient (CCC)=0.97]. In the external TCIA dataset, the model had higher reliability than R1 and R2 on full vs reduced dose CT scans [model mean (SD) DSC: 0.96 (0.02), CCC = 0.995 vs R1 DSC: 0.83 (0.07), CCC = 0.89, and R2 DSC:0.87 (0.04), CCC = 0.97]. The DSC and volume concordance correlations for R1 vs R2 (inter-reader reliability) were 0.85 (0.07), CCC = 0.90 at full dose and 0.83 (0.07), CCC = 0.96 at reduced dose datasets. There was good reliability between model and R1 at both full and reduced dose CT [full dose: DSC: 0.81 (0.07), CCC = 0.83 and reduced dose DSC:0.81 (0.08), CCC = 0.87]. Likewise, there was good reliability between model and R2 at both full and reduced dose CT [full dose: DSC: 0.84 (0.05), CCC = 0.89 and reduced dose DSC:0.83(0.06), CCC = 0.89]. There was no difference in model-predicted and GT pancreatic volume in TCIA dataset (mean predicted volume 96 cc (33); GT pancreatic volume 89 cc (30), p = 0.31). Model had mean (SD) DSC: 0.89 (0.04) (minimum-maximum DSC: 0.79 -0.96) on the NIH-PCT dataset., Conclusion: A 3D CNN developed on the largest dataset of CTs is accurate for fully automated volumetric pancreas segmentation and is generalizable across a wide range of CT slice thicknesses, radiation dose, and patient gender. This 3D CNN offers a scalable tool to leverage biomarkers from pancreas morphometrics and radiomics for pancreatic diseases including for early pancreatic cancer detection., (© 2021 American Association of Physicists in Medicine.)

Purpose: Organ segmentation of computed tomography (CT) imaging is essential for radiotherapy treatment planning. Treatment planning requires segmentation not only of the affected tissue, but nearby healthy organs-at-risk, which is laborious and time-consuming. We present a fully automated segmentation method based on the three-dimensional (3D) U-Net convolutional neural network (CNN) capable of whole abdomen and pelvis segmentation into 33 unique organ and tissue structures, including tissues that may be overlooked by other automated segmentation approaches such as adipose tissue, skeletal muscle, and connective tissue and vessels. Whole abdomen segmentation is capable of quantifying exposure beyond a handful of organs-at-risk to all tissues within the abdomen., Methods: Sixty-six (66) CT examinations of 64 individuals were included in the training and validation sets and 18 CT examinations from 16 individuals were included in the test set. All pixels in each examination were segmented by image analysts (with physician correction) and assigned one of 33 labels. Segmentation was performed with a 3D U-Net variant architecture which included residual blocks, and model performance was quantified on 18 test cases. Human interobserver variability (using semiautomated segmentation) was also reported on two scans, and manual interobserver variability of three individuals was reported on one scan. Model performance was also compared to several of the best models reported in the literature for multiple organ segmentation., Results: The accuracy of the 3D U-Net model ranges from a Dice coefficient of 0.95 in the liver, 0.93 in the kidneys, 0.79 in the pancreas, 0.69 in the adrenals, and 0.51 in the renal arteries. Model accuracy is within 5% of human segmentation in eight of 19 organs and within 10% accuracy in 13 of 19 organs., Conclusions: The CNN approaches the accuracy of human tracers and on certain complex organs displays more consistent prediction than human tracers. Fully automated deep learning-based segmentation of CT abdomen has the potential to improve both the speed and accuracy of radiotherapy dose prediction for organs-at-risk., (© 2020 American Association of Physicists in Medicine.)

Purpose: Early identification of ischemic stroke plays a significant role in treatment and potential recovery of damaged brain tissue. In noncontrast CT (ncCT), the differences between ischemic changes and healthy tissue are usually very subtle during the hyperacute phase (< 8 h from the stroke onset). Therefore, visual comparison of both hemispheres is an important step in clinical assessment. A quantitative symmetry-based analysis of texture features of ischemic lesions in noncontrast CT images may provide an important information for differentiation of ischemic and healthy brain tissue in this phase., Methods: One hundred thirty-nine (139) ncCT scans of hyperacute ischemic stroke with follow-up magnetic resonance diffusion-weighted (MR-DW) images were collected. The regions of stroke were identified in the MR-DW images, which were spatially aligned to corresponding ncCT images. A state-of-the-art symmetric diffeomorphic image registration was utilized for the alignment of CT and MR-DW, for identification of individual brain hemispheres, and for localization of the region representing healthy tissue contralateral to the stroke cores. Texture analysis included extraction and classification of co-occurrence and run-length texture-based image features in the regions of ischemic stroke and their contralateral regions., Results: The classification schemes achieved area under the receiver operating characteristic [Az] ≈ 0.82 for the whole dataset. There was no statistically significant difference in the performance of classifiers for the data sets with time between 2 and 8 hours from symptom onset. The performance of the classifiers did not depend on the size of the stroke regions., Conclusions: The results provide a set of optimal texture features which are suitable for distinguishing between hyperacute ischemic lesions and their corresponding contralateral brain tissue in noncontrast CT. This work is an initial step toward development of an automated decision support system for detection of hyperacute ischemic stroke lesions on noncontrast CT of the brain., (© 2016 American Association of Physicists in Medicine.)

Purpose: Imaging biomarker research focuses on discovering relationships between radiological features and histological findings. In glioblastoma patients, methylation of the O(6)-methylguanine methyltransferase (MGMT) gene promoter is positively correlated with an increased effectiveness of current standard of care. In this paper, the authors investigate texture features as potential imaging biomarkers for capturing the MGMT methylation status of glioblastoma multiforme (GBM) tumors when combined with supervised classification schemes., Methods: A retrospective study of 155 GBM patients with known MGMT methylation status was conducted. Co-occurrence and run length texture features were calculated, and both support vector machines (SVMs) and random forest classifiers were used to predict MGMT methylation status., Results: The best classification system (an SVM-based classifier) had a maximum area under the receiver-operating characteristic (ROC) curve of 0.85 (95% CI: 0.78-0.91) using four texture features (correlation, energy, entropy, and local intensity) originating from the T2-weighted images, yielding at the optimal threshold of the ROC curve, a sensitivity of 0.803 and a specificity of 0.813., Conclusions: Results show that supervised machine learning of MRI texture features can predict MGMT methylation status in preoperative GBM tumors, thus providing a new noninvasive imaging biomarker.

Purpose: Primary goal of this study is to select optimal registration schemes in the framework of interstitial lung disease (ILD) follow-up analysis in CT., Methods: A set of 128 multiresolution schemes composed of multiresolution nonrigid and combinations of rigid and nonrigid registration schemes are evaluated, utilizing ten artificially warped ILD follow-up volumes, originating from ten clinical volumetric CT scans of ILD affected patients, to select candidate optimal schemes. Specifically, all combinations of four transformation models (three rigid: rigid, similarity, affine and one nonrigid: third order B-spline), four cost functions (sum-of-square distances, normalized correlation coefficient, mutual information, and normalized mutual information), four gradient descent optimizers (standard, regular step, adaptive stochastic, and finite difference), and two types of pyramids (recursive and Gaussian-smoothing) were considered. The selection process involves two stages. The first stage involves identification of schemes with deformation field singularities, according to the determinant of the Jacobian matrix. In the second stage, evaluation methodology is based on distance between corresponding landmark points in both normal lung parenchyma (NLP) and ILD affected regions. Statistical analysis was performed in order to select near optimal registration schemes per evaluation metric. Performance of the candidate registration schemes was verified on a case sample of ten clinical follow-up CT scans to obtain the selected registration schemes., Results: By considering near optimal schemes common to all ranking lists, 16 out of 128 registration schemes were initially selected. These schemes obtained submillimeter registration accuracies in terms of average distance errors 0.18 ± 0.01 mm for NLP and 0.20 ± 0.01 mm for ILD, in case of artificially generated follow-up data. Registration accuracy in terms of average distance error in clinical follow-up data was in the range of 1.985-2.156 mm and 1.966-2.234 mm, for NLP and ILD affected regions, respectively, excluding schemes with statistically significant lower performance (Wilcoxon signed-ranks test, p < 0.05), resulting in 13 finally selected registration schemes., Conclusions: Selected registration schemes in case of ILD CT follow-up analysis indicate the significance of adaptive stochastic gradient descent optimizer, as well as the importance of combined rigid and nonrigid schemes providing high accuracy and time efficiency. The selected optimal deformable registration schemes are equivalent in terms of their accuracy and thus compatible in terms of their clinical outcome.

Accurate and automated lung field (LF) segmentation in high-resolution computed tomography (HRCT) is highly challenged by the presence of pathologies affecting lung borders, also affecting the performance of computer-aided diagnosis (CAD) schemes. In this work, a two-dimensional LF segmentation algorithm adapted to interstitial pneumonia (IP) patterns is presented. The algorithm employs k-means clustering followed by a filling operation to obtain an initial LF order estimate. The final LF border is obtained by an iterative support vector machine neighborhood labeling of border pixels based on gray level and wavelet coefficient statistics features. A second feature set based on gray level averaging and gradient features was also investigated to evaluate its effect on segmentation performance of the proposed method. The proposed method is evaluated on a dataset of 22 HRCT cases spanning a range of IP patterns such as ground glass, reticular, and honeycombing. The accuracy of the method is assessed using area overlap and shape differentiation metrics (d(mean), d(rms), and d(max)), by comparing automatically derived lung borders to manually traced ones, and further compared to a gray level thresholding-based (GLT-based) method. Accuracy of the methods evaluated is also compared to interobserver variability. The proposed method incorporating gray level and wavelet coefficient statistics demonstrated the highest segmentation accuracy, averaged over left and right LFs (overlap=0.954, d(mean)=1.080 mm, d(rms)=1.407 mm, and d(max)=4.944 mm), which is statistically significant (two-tailed student's t test for paired data, p<0.0083) with respect to all metrics considered as compared to the proposed method incorporating gray level averaging and gradient features (overlap=0.918, d(mean)=2.354 mm, d(rms)=3.711 mm, and d(max)=14.412 mm) and the GLT-based method (overlap=0.897, d(mean)=3.618 mm, d(rms)=5.007 mm, and d(max)=16.893 mm). The performance of the three segmentation methods, although decreased as IP pattern severity level (mild, moderate, and severe) was increased, did not demonstrate statistically significant difference (two-tailed student's t test for unpaired data, p>0.0167 for all metrics considered). Finally, the accuracy of the proposed method, based on gray level and wavelet coefficient statistics ranges within interobserver variability. The proposed segmentation method could be used as an initial stage of a CAD scheme for IP patterns.