Richard Castillo

A framework for evaluation of deformable image registration spatial accuracy using large landmark point sets.

Expert landmark correspondences are widely reported for evaluating deformable image registration (DIR) spatial accuracy. In this report, we present a framework for objective evaluation of DIR spatial accuracy using large sets of expert-determined landmark point pairs. Large samples (>1100) of pulmonary landmark point pairs were manually generated for five cases. Estimates of inter- and intra-observer variation were determined from repeated registration. Comparative evaluation of DIR spatial accuracy was performed for two algorithms, a gradient-based optical flow algorithm and a landmark-based moving least-squares algorithm. The uncertainty of spatial error estimates was found to be inversely proportional to the square root of the number of landmark point pairs and directly proportional to the standard deviation of the spatial errors. Using the statistical properties of this data, we performed sample size calculations to estimate the average spatial accuracy of each algorithm with 95% confidence intervals within a 0.5 mm range. For the optical flow and moving least-squares algorithms, the required sample sizes were 1050 and 36, respectively. Comparative evaluation based on fewer than the required validation landmarks results in misrepresentation of the relative spatial accuracy. This study demonstrates that landmark pairs can be used to assess DIR spatial accuracy within a narrow uncertainty range.

Implementation and evaluation of various demons deformable image registration algorithms on a GPU

Online adaptive radiation therapy (ART) promises the ability to deliver an optimal treatment in response to daily patient anatomic variation. A major technical barrier for the clinical implementation of online ART is the requirement of rapid image segmentation. Deformable image registration (DIR) has been used as an automated segmentation method to transfer tumor/organ contours from the planning image to daily images. However, the current computational time of DIR is insufficient for online ART. In this work, this issue is addressed by using computer graphics processing units (GPUs). A gray-scale-based DIR algorithm called demons and five of its variants were implemented on GPUs using the compute unified device architecture (CUDA) programming environment. The spatial accuracy of these algorithms was evaluated over five sets of pulmonary 4D CT images with an average size of 256 x 256 x 100 and more than 1100 expert-determined landmark point pairs each. For all the testing scenarios presented in this paper, the GPU-based DIR computation required around 7 to 11 s to yield an average 3D error ranging from 1.5 to 1.8 mm. It is interesting to find out that the original passive force demons algorithms outperform subsequently proposed variants based on the combination of accuracy, efficiency and ease of implementation.

Four-dimensional deformable image registration using trajectory modeling

A four-dimensional deformable image registration (4D DIR) algorithm, referred to as 4D local trajectory modeling (4DLTM), is presented and applied to thoracic 4D computed tomography (4DCT) image sets. The theoretical framework on which this algorithm is built exploits the incremental continuity present in 4DCT component images to calculate a dense set of parameterized voxel trajectories through space as functions of time. The spatial accuracy of the 4DLTM algorithm is compared with an alternative registration approach in which component phase to phase (CPP) DIR is utilized to determine the full displacement between maximum inhale and exhale images. A publically available DIR reference database (http://www.dir-lab.com) is utilized for the spatial accuracy assessment. The database consists of ten 4DCT image sets and corresponding manually identified landmark points between the maximum phases. A subset of points are propagated through the expiratory 4DCT component images. Cubic polynomials were found to provide sufficient flexibility and spatial accuracy for describing the point trajectories through the expiratory phases. The resulting average spatial error between the maximum phases was 1.25 mm for the 4DLTM and 1.44 mm for the CPP. The 4DLTM method captures the long-range motion between 4DCT extremes with high spatial accuracy.

Ventilation from four-dimensional computed tomography: density versus Jacobian methods

Two calculation methods to produce ventilation images from four-dimensional computed tomography (4DCT) acquired without added contrast have been reported. We reported a method to obtain ventilation images using deformable image registration (DIR) and the underlying CT density information. A second method performs the ventilation image calculation from the DIR result alone, using the Jacobian determinant of the deformation field to estimate the local volume changes resulting from ventilation. For each of these two approaches, there are variations on their implementation. In this study, two implementations of the Jacobian-based methodology are evaluated, as well as a single density change-based model for calculating the physiologic specific ventilation from 4DCT. In clinical practice, (99m)Tc-labeled aerosol single photon emission computed tomography (SPECT) is the standard method used to obtain ventilation images in patients. In this study, the distributions of ventilation obtained from the CT-based ventilation image calculation methods are compared with those obtained from the clinical standard SPECT ventilation imaging. Seven patients with 4DCT imaging and standard (99m)Tc-labeled aerosol SPECT/CT ventilation imaging obtained on the same day as part of a prospective validation study were selected. The results of this work demonstrate the equivalence of the Jacobian-based methodologies for quantifying the specific ventilation on a voxel scale. Additionally, we found that both Jacobian- and density-change-based methods correlate well with global measurements of the resting tidal volume. Finally, correlation with the clinical SPECT was assessed using the Dice similarity coefficient, which showed statistically higher (p-value < 10(-4)) correlation between density-change-based specific ventilation and the clinical reference than did either Jacobian-based implementation.

Use of 4-dimensional computed tomography-based ventilation imaging to correlate lung dose and function with clinical outcomes

PURPOSE Four-dimensional computed tomography (4DCT)-based ventilation is an emerging imaging modality that can be used in the thoracic treatment planning process. The clinical benefit of using ventilation images in radiation treatment plans remains to be tested. The purpose of the current work was to test the potential benefit of using ventilation in treatment planning by evaluating whether dose to highly ventilated regions of the lung resulted in increased incidence of clinical toxicity. METHODS AND MATERIALS Pretreatment 4DCT data were used to compute pretreatment ventilation images for 96 lung cancer patients. Ventilation images were calculated using 4DCT data, deformable image registration, and a density-change based algorithm. Dose-volume and ventilation-based dose function metrics were computed for each patient. The ability of the dose-volume and ventilation-based dose-function metrics to predict for severe (grade 3+) radiation pneumonitis was assessed using logistic regression analysis, area under the curve (AUC) metrics, and bootstrap methods. RESULTS A specific patient example is presented that demonstrates how incorporating ventilation-based functional information can help separate patients with and without toxicity. The logistic regression significance values were all lower for the dose-function metrics (range P=.093-.250) than for their dose-volume equivalents (range, P=.331-.580). The AUC values were all greater for the dose-function metrics (range, 0.569-0.620) than for their dose-volume equivalents (range, 0.500-0.544). Bootstrap results revealed an improvement in model fit using dose-function metrics compared to dose-volume metrics that approached significance (range, P=.118-.155). CONCLUSIONS To our knowledge, this is the first study that attempts to correlate lung dose and 4DCT ventilation-based function to thoracic toxicity after radiation therapy. Although the results were not significant at the .05 level, our data suggests that incorporating ventilation-based functional imaging can improve prediction for radiation pneumonitis. We present an important first step toward validating the use of 4DCT-based ventilation imaging in thoracic treatment planning.

Lung Texture in Serial Thoracic Computed Tomography Scans: Correlation of Radiomics-based Features With Radiation Therapy Dose and Radiation Pneumonitis Development

PURPOSE To assess the relationship between radiation dose and change in a set of mathematical intensity- and texture-based features and to determine the ability of texture analysis to identify patients who develop radiation pneumonitis (RP). METHODS AND MATERIALS A total of 106 patients who received radiation therapy (RT) for esophageal cancer were retrospectively identified under institutional review board approval. For each patient, diagnostic computed tomography (CT) scans were acquired before (0-168 days) and after (5-120 days) RT, and a treatment planning CT scan with an associated dose map was obtained. 32- × 32-pixel regions of interest (ROIs) were randomly identified in the lungs of each pre-RT scan. ROIs were subsequently mapped to the post-RT scan and the planning scan dose map by using deformable image registration. The changes in 20 feature values (ΔFV) between pre- and post-RT scan ROIs were calculated. Regression modeling and analysis of variance were used to test the relationships between ΔFV, mean ROI dose, and development of grade ≥2 RP. Area under the receiver operating characteristic curve (AUC) was calculated to determine each features ability to distinguish between patients with and those without RP. A classifier was constructed to determine whether 2- or 3-feature combinations could improve RP distinction. RESULTS For all 20 features, a significant ΔFV was observed with increasing radiation dose. Twelve features changed significantly for patients with RP. Individual texture features could discriminate between patients with and those without RP with moderate performance (AUCs from 0.49 to 0.78). Using multiple features in a classifier, AUC increased significantly (0.59-0.84). CONCLUSIONS A relationship between dose and change in a set of image-based features was observed. For 12 features, ΔFV was significantly related to RP development. This study demonstrated the ability of radiomics to provide a quantitative, individualized measurement of patient lung tissue reaction to RT and assess RP development.

Use of weekly 4DCT-based ventilation maps to quantify changes in lung function for patients undergoing radiation therapy

PURPOSE A method has been proposed to calculate ventilation maps from four-dimensional computed tomography (4DCT) images. Weekly 4DCT data were acquired throughout the course of radiation therapy for patients with lung cancer. The purpose of our work was to use ventilation maps calculated from weekly 4DCT data to study how ventilation changed throughout radiation therapy. METHODS Quantitative maps representing ventilation were generated for six patients. Deformable registration was used to link corresponding lung volume elements between the inhale and exhale phases of the 4DCT dataset. Following spatial registration, corresponding Hounsfield units were input into a density-change-based model for quantifying the local ventilation. The ventilation data for all weeks were registered to the pretreatment ventilation image set. We quantitatively analyzed the data by defining regions of interest (ROIs) according to dose (V(20)) and lung lobe and by tracking the weekly ventilation of each ROI throughout treatment. The slope of the linear fit to the weekly ventilation data was used to evaluate the change in ventilation throughout treatment. A positive slope indicated an increase in ventilation, a negative slope indicated a decrease in ventilation, and a slope of 0 indicated no change. The dose ROI ventilation and slope data were used to study how ventilation changed throughout treatment as a function of dose. The lung lobe ROI ventilation data were used to study the impact of the presence of tumor on pretreatment ventilation. In addition, the lobe ROI data were used to study the impact of tumor reduction on ventilation change throughout treatment. RESULTS Using the dose ROI data, we found that three patients had an increase in weekly ventilation as a function of dose (slopes of 1.1, 1.4, and 1.5) and three patients had no change or a slight decrease in ventilation as a function of dose (slopes of 0.3, -0.6, -0.5). Visually, pretreatment ventilation appeared to be lower in the lobes that contained tumor. Pretreatment ventilation was 39% for lobes that contained tumor and 54% for lobes that did not contain tumor. The difference in ventilation between the two groups was statistically significant (p = 0.017). When the weekly lobe ventilation data were qualitatively observed, two distinct patterns emerged. When the tumor volume in a lobe was reduced, ventilation increased in the lobe. When the tumor volume was not reduced, the ventilation distribution did not change. The average slope of the group of lobes that contained tumors that shrank was 1.18, while the average slope of the group that did not contain tumors (or contained tumors that did not shrink) was -0.32. The slopes for the two groups were significantly different (p = 0.014). CONCLUSIONS We did not find a consistent pattern of ventilation change as a function of radiation dose. Pretreatment ventilation was significantly lower for lobes that contained tumor, due to occlusion of the central airway. The weekly lobe ventilation data indicated that when tumor volume shrinks, ventilation increases, and when the thoracic anatomy is not visibly changed, ventilation is likely to remain unchanged.

A reference dataset for deformable image registration spatial accuracy evaluation using the COPDgene study archive

Landmark point-pairs provide a strategy to assess deformable image registration (DIR) accuracy in terms of the spatial registration of the underlying anatomy depicted in medical images. In this study, we propose to augment a publicly available database (www.dir-lab.com) of medical images with large sets of manually identified anatomic feature pairs between breath-hold computed tomography (BH-CT) images for DIR spatial accuracy evaluation. Ten BH-CT image pairs were randomly selected from the COPDgene study cases. Each patient had received CT imaging of the entire thorax in the supine position at one-fourth dose normal expiration and maximum effort full dose inspiration. Using dedicated in-house software, an imaging expert manually identified large sets of anatomic feature pairs between images. Estimates of inter- and intra-observer spatial variation in feature localization were determined by repeat measurements of multiple observers over subsets of randomly selected features. 7298 anatomic landmark features were manually paired between the 10 sets of images. Quantity of feature pairs per case ranged from 447 to 1172. Average 3D Euclidean landmark displacements varied substantially among cases, ranging from 12.29 (SD: 6.39) to 30.90 (SD: 14.05) mm. Repeat registration of uniformly sampled subsets of 150 landmarks for each case yielded estimates of observer localization error, which ranged in average from 0.58 (SD: 0.87) to 1.06 (SD: 2.38) mm for each case. The additions to the online web database (www.dir-lab.com) described in this work will broaden the applicability of the reference data, providing a freely available common dataset for targeted critical evaluation of DIR spatial accuracy performance in multiple clinical settings. Estimates of observer variance in feature localization suggest consistent spatial accuracy for all observers across both four-dimensional CT and COPDgene patient cohorts.

Spatial correspondence of 4D CT ventilation and SPECT pulmonary perfusion defects in patients with malignant airway stenosis

To determine the spatial overlap agreement between four-dimensional computed tomography (4D CT) ventilation and single photon emission computed tomography (SPECT) perfusion hypo-functioning pulmonary defect regions in a patient population with malignant airway stenosis. Treatment planning 4D CT images were obtained retrospectively for ten lung cancer patients with radiographically demonstrated airway obstruction due to gross tumor volume. Each patient also received a SPECT perfusion study within one week of the planning 4D CT, and prior to the initiation of treatment. Deformable image registration was used to map corresponding lung tissue elements between the extreme component phase images, from which quantitative three-dimensional (3D) images representing the local pulmonary specific ventilation were constructed. Semi-automated segmentation of the percentile perfusion distribution was performed to identify regional defects distal to the known obstructing lesion. Semi-automated segmentation was similarly performed by multiple observers to delineate corresponding defect regions depicted on 4D CT ventilation. Normalized Dice similarity coefficient (NDSC) indices were determined for each observer between SPECT perfusion and 4D CT ventilation defect regions to assess spatial overlap agreement. Tidal volumes determined from 4D CT ventilation were evaluated versus measurements obtained from lung parenchyma segmentation. Linear regression resulted in a linear fit with slope = 1.01 (R² = 0.99). Respective values for the average DSC, NDSC(1 mm) and NDSC(2 mm) for all cases and multiple observers were 0.78, 0.88 and 0.99, indicating that, on average, spatial overlap agreement between ventilation and perfusion defect regions was comparable to the threshold for agreement within 1-2 mm uncertainty. Corresponding coefficients of variation for all metrics were similarly in the range: 0.10%-19%. This study is the first to quantitatively assess 3D spatial overlap agreement between clinically acquired SPECT perfusion and specific ventilation from 4D CT. Results suggest high correlation between methods within the sub-population of lung cancer patients with malignant airway stenosis.

[18F]-FDG uptake dose-response correlates with radiation pneumonitis in lung cancer patients

PURPOSE To quantify the post-radiotherapy 2-[(18)F]-fluoro-2-deoxyglucose (FDG) pulmonary uptake dose-response in lung cancer patients and determine its relationship with radiation pneumonitis symptoms. METHODS AND MATERIALS The data from 24 patients treated for lung cancer with thoracic radiotherapy who received restaging PET/CT imaging between 4 and 12 weeks after radiotherapy completion were evaluated. Their radiation dose distribution was registered with the post-treatment restaging PET/CT. Using histogram analysis, the voxel average FDG-PET uptake vs. radiation dose was obtained for each case and linear regression was performed. The resulting slope, the pulmonary metabolic radiation response (PMRR), was used to characterize the dose-response. The Common Toxicity Criteria version 3 was used to score clinical pulmonary toxicity symptoms. Receiver operating characteristic (ROC) curves were used to determine the level of FDG uptake vs. dose, MLD, V(5), V(10), V(20), and V(30) that can best predict symptomatic and asymptomatic patients. RESULTS The median time between radiotherapy completion and FDG-PET imaging was 59 days (range, 26-70 days). The median of the mean SUV from lung that received 0-5 Gy was 1.00 (range, 0.37-1.48), 5-10 Gy was 1.01 (range, 0.37-1.77), 10-20 Gy was 1.04 (0.42-1.53), and >20 Gy was 1.29 (range, 0.41-8.01). Using the dose range of 0 Gy to the maximum dose minus 10 Gy, hierarchical linear regression model of the radiation dose and normalized FDG uptake per case found an adequate fit with the linear model. Pneumonitis scores were: Grade 0 for 13, Grade 1 for 5, Grade 2 for 6, and Grade 3, 4 or 5 for none. Using a PMRR threshold of 0.017 yields an associated true positive rate of 0.67 and false positive rate of 0.15 with average error of 30%. A V(5) threshold of 57.6 gives an associated true positive rate of 0.67 and false positive rate of 0.05 with a 20% average error. CONCLUSION The metabolic radiation pneumonitis dose-response was evaluated from post-treatment FDG-PET/CT imaging. Statistical modeling found a linear relationship. The FDG uptake dose-response and V(5) correlated with symptomatic radiation pneumonitis.