Pulmonary nodule registration: rigid or nonrigid?

Re-Identification and growth detection of pulmonary nodules without image registration using 3D siamese neural networks

Lung cancer follow-up is a complex, error prone, and time consuming task for clinical radiologists. Several lung CT scan images taken at different time points of a given patient need to be individually inspected, looking for possible cancerogenous nodules. Radiologists mainly focus their attention in nodule size, density, and growth to assess the existence of malignancy. In this study, we present a novel method based on a 3D siamese neural network, for the re-identification of nodules in a pair of CT scans of the same patient without the need for image registration. The network was integrated into a two-stage automatic pipeline to detect, match, and predict nodule growth given pairs of CT scans. Results on an independent test set reported a nodule detection sensitivity of 94.7%, an accuracy for temporal nodule matching of 88.8%, and a sensitivity of 92.0% with a precision of 88.4% for nodule growth detection.

Correlation of normal lung density changes with dose after stereotactic body radiotherapy (SBRT) for early stage lung cancer

BACKGROUND AND PURPOSE

To investigate the correlation between normal lung CT density changes with dose accuracy and outcome after stereotactic body radiation therapy (SBRT) for patients with early stage non-small-cell lung cancer (NSCLC).

MATERIALS AND METHODS

Thirty-one patients (with a total of 33 lesions) with non-small cell lung cancer were selected out of 270 patients treated with SBRT at a single institution between 2003 and 2009. Out of these 31 patients, 10 patients had developed radiation pneumonitis (RP). Dose distributions originally planned using a 1-D pencil beam-based dose algorithm were retrospectively recomputed using different algorithms. Prescription dose was 48 Gy in 4 fractions in most patients. Planning CT images were rigidly registered to follow-up CT datasets at 3-9 months after treatment. Corresponding dose distributions were mapped from planning to follow-up CT images. Hounsfield Unit (HU) changes in lung density in individual, 5 Gy, dose bins from 5 to 45 Gy were assessed in the peri-tumoral region. Correlations between HU changes in various normal lung regions, dose indices (V20, MLD, generalized equivalent uniform dose (gEUD)), and RP grade were investigated.

RESULTS

Strong positive correlation was found between HU changes in the peri-tumoral region and RP grade (Spearman's r = 0.760; p < 0.001). Positive correlation was also observed between RP and HU changes in the region covered by V20 for all algorithms (Spearman's r ≥ 0.738; p < 0.001). Additionally, V20, MLD, and gEUD were significantly correlated with RP grade (p < 0.01). MLD in the peri-tumoral region computed with model-based algorithms was 5-7% lower than the PB-based methods.

CONCLUSION

Changes of lung density in the peri-tumoral lung and in the region covered by V20 were strongly associated with RP grade. Relative to model-based methods, PB algorithms over-estimated mean peri-tumoral dose and showed displacement of the high-dose region, which correlated with HU changes on follow-up CT scans.

Preliminary Assessment of an Optical Flow Method (OFM) for Nonrigid Registration and Temporal Subtraction (TS) of Serial CT Examinations to Facilitate Evaluation of Interval Change in Metastatic Lung Nodules

RATIONALE AND OBJECTIVES

Accurate assessment of size change of lung nodules on chest computed tomography (CT) is important for diagnosis and response assessment. However, manual methods are time-consuming and error-prone. We therefore assessed whether an optical flow method (OFM) with temporal subtraction (TS) can facilitate detection and quantification of lung nodule change on serial CT datasets.

MATERIALS AND METHODS

Serial chest CT examinations were selected from 12 patients with multiple pulmonary metastases. Lung nodules were evaluated for change in size using: (1) OFM with TS and (2) reference standard visual and manual assessment. Average time required to assess interval change using both methods was recorded and compared. Concordance of agreement between OFM with TS and reference standard assessment for nodule change was examined.

RESULTS

285 solid pulmonary nodules were evaluated. The average time per nodule to assess interval change in nodule size by OFM with TS (mean 1.15 + 0.5 minutes) was significantly less (P = 0.02) than that the reference standard approach (mean 1.56 + 0.5 minutes). Agreement between OFM with TS and reference standard occurred for 63.2% of nodules overall (kappa = 0.50, standard error 0.35, P< 0.00001), and significantly increased with larger nodule size (kappa = 0.48 for nodules <5 mm; kappa = 0.94 for nodules >20 mm, P < 0.0001).

CONCLUSIONS

This preliminary study demonstrates the feasibility of an OFM with TS to assess for interval change in metastatic lung nodules on serial CT examinations with significantly improved reading speed and moderate agreement relative to reference standard assessment. Agreement improved with larger nodule size.

Modified Quality Threshold Clustering for Temporal Analysis and Classification of Lung Lesions

Lung cancer is the type of cancer that most often kills after the initial diagnosis. To aid the specialist in its diagnosis, temporal evaluation is a potential tool for analyzing indeterminate lesions, which may be benign or malignant, during treatment. With this goal in mind, a methodology is herein proposed for the analysis, quantification, and visualization of changes in lung lesions. This methodology uses a modified version of the quality threshold clustering algorithm to associate each voxel of the lesion to a cluster, and changes in the lesion over time are defined in terms of voxel moves to another cluster. In addition, statistical features are extracted for classification of the lesion as benign or malignant. To develop the proposed methodology, two databases of pulmonary lesions were used, one for malignant lesions in treatment (public) and the other for indeterminate cases (private). We determined that the density change percentage varied from 6.22% to 36.93% of lesion volume in the public database of malignant lesions under treatment and from 19.98% to 38.81% in the private database of lung nodules. Additionally, other inter-cluster density change measures were obtained. These measures indicate the degree of change in the clusters and how each of them is abundant in relation to volume. From the statistical analysis of regions in which the density changes occurred, we were able to discriminate lung lesions with an accuracy of 98.41%, demonstrating that these changes can indicate the true nature of the lesion. In addition to visualizing the density changes occurring in lesions over time, we quantified these changes and analyzed the entire set through volumetry, which is the technique most commonly used to analyze changes in pulmonary lesions.

Statistical tools for the temporal analysis and classification of lung lesions

BACKGROUND AND OBJECTIVE

Lung cancer remains one of the most common cancers globally. Temporal evaluation is an important tool for analyzing the malignant behavior of lesions during treatment, or of indeterminate lesions that may be benign. This work proposes a methodology for the analysis, quantification, and visualization of small (local) and large (global) changes in lung lesions. In addition, we extract textural features for the classification of lesions as benign or malignant.

METHODS

We employ the statistical concept of uncertainty to associate each voxel of a lesion to a probability that changes occur in the lesion over time. We employ the Jensen divergence and hypothesis test locally to verify voxel-to-voxel changes, and globally to capture changes in lesion volumes.

RESULTS

For the local hypothesis test, we determine that the change in density varies by between 3.84 and 40.01% of the lesion volume in a public database of malignant lesions under treatment, and by between 5.76 and 35.43% in a private database of benign lung nodules. From the texture analysis of regions in which the density changes occur, we are able to discriminate lung lesions with an accuracy of 98.41%, which shows that these changes can indicate the true nature of the lesion.

CONCLUSION

In addition to the visual aspects of the density changes occurring in the lesions over time, we quantify these changes and analyze the entire set using volumetry. In the case of malignant lesions, large b-divergence values are associated with major changes in lesion volume. In addition, this occurs when the change in volume is small but is associated with significant changes in density, as indicated by the histogram divergence. For benign lesions, the methodology shows that even in cases where the change in volume is small, a change of density occurs. This proves that even in lesions that appear stable, a change in density occurs.

Selecting registration schemes in case of interstitial lung disease follow-up in CT

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.

Correlation of Clinical and Dosimetric Parameters With Radiographic Lung Injury Following Stereotactic Body Radiotherapy

Radiographic changes occur in over half of patients treated with stereotactic body radiotherapy (SBRT) to the lung, correlating histopathologically with injury. We quantified radiographic density changes (ie, fibrosis) at 3, 6, and 12 months and investigated the relationship between these volumes and clinical and dosimetric parameters. The study population consisted of patients treated with SBRT to the lung for stage I primary lung cancers (n = 39) or oligometastatic lesions (n = 17). Fractionation schemes included 3 fractions of 12, 14, or 18 gray (Gy) and 4 fractions of 12 or 12.5 Gy prescribed to cover 95% of the planning target volume (PTV). Planning computed tomography (CT) scans were rigidly registered to follow-up CT scans obtained at intervals of 3, 6, and 12 months. Fibrotic volumes were contoured on the follow-up scans. Associations between the volume of fibrosis and clinical and dosimetric parameters were investigated using univariate linear regression. Scans were available for 65 and 47 lesions at 6 and 12 months, respectively. Age, years since quitting smoking, and GOLD Global Initiative for Chronic Obstructive Lung Disease score were significantly associated with increasing volume of fibrosis (P < .05). Total dose, dose per fraction, PTV, and volumetric parameters (V0-V55) were also significantly associated with increasing volumes of fibrosis (P < .01). For dosimetric parameters, the effect was largest for V55. Age, significant smoking history, and GOLD score were significantly associated with increasing volumes of fibrosis following SBRT. In a multivariate model adjusted for age and smoking history, V10 through V50 and PTV size remained significant predictors of fibrotic volume. Further, there is a strong dose-response relationship between the volume of lung exposed to a certain dose and the fibrotic volume. The predominant kinetic patterns of fibrosis demonstrate peaking fibrotic volumes at 6 and 12 months. These results provide insight for expectations of fibrosis after SBRT.

Estimation of the uncertainty of elastic image registration with the demons algorithm

The accuracy of elastic image registration is limited. We propose an approach to detect voxels where registration based on the demons algorithm is likely to perform inaccurately, compared to other locations of the same image. The approach is based on the assumption that the local reproducibility of the registration can be regarded as a measure of uncertainty of the image registration. The reproducibility is determined as the standard deviation of the displacement vector components obtained from multiple registrations. These registrations differ in predefined initial deformations. The proposed approach was tested with artificially deformed lung images, where the ground truth on the deformation is known. In voxels where the result of the registration was less reproducible, the registration turned out to have larger average registration errors as compared to locations of the same image, where the registration was more reproducible. The proposed method can show a clinician in which area of the image the elastic registration with the demons algorithm cannot be expected to be accurate.