Validation of a Model-Based Segmentation Approach to Propagating Normal Anatomic Regions of Interest Through the 10 Phases of Respiration

Deterministic small-scale undulations of image-based risk predictions from the deep learning of lung tumors in motion

INTRODUCTION

Deep learning (DL) models that use medical images to predict clinical outcomes are poised for clinical translation. For tumors that reside in organs that move, however, the impact of motion (i.e., degenerated object appearance or blur) on DL model accuracy remains unclear. We examine the impact of tumor motion on an image-based DL framework that predicts local failure risk after lung stereotactic body radiotherapy (SBRT).

METHODS

We input pre-therapy free breathing (FB) computed tomography (CT) images from 849 patients treated with lung SBRT into a multitask deep neural network to generate an image fingerprint signature (or DL score) that predicts time-to-event local failure outcomes. The network includes a convolutional neural network encoder for extracting imaging features and building a task-specific fingerprint, a decoder for estimating handcrafted radiomic features, and a task-specific network for generating image signature for radiotherapy outcome prediction. The impact of tumor motion on the DL scores was then examined for a holdout set of 468 images from 39 patients comprising: (1) FB CT, (2) four-dimensional (4D) CT, and (3) maximum-intensity projection (MIP) images. Tumor motion was estimated using a 3D vector of the maximum distance traveled, and its association with DL score variance was assessed by linear regression.

FINDINGS

The variance and amplitude in 4D CT image-derived DL scores were associated with tumor motion (R²  = 0.48 and 0.46, respectively). Specifically, DL score variance was deterministic and represented by sinusoidal undulations in phase with the respiratory cycle. DL scores, but not tumor volumes, peaked near end-exhalation. The mean of the scores derived from 4D CT images and the score obtained from FB CT images were highly associated (Pearson r = 0.99). MIP-derived DL scores were significantly higher than 4D- or FB-derived risk scores (p < 0.0001).

INTERPRETATION

An image-based DL risk score derived from a series of 4D CT images varies in a deterministic, sinusoidal trajectory in a phase with the respiratory cycle. These results indicate that DL models of tumors in motion can be robust to fluctuations in object appearance due to movement and can guide standardization processes in the clinical translation of DL models for patients with lung cancer.

Validation of Automatic Contour Propagation for 4D Treatment Planning Using Multiple Metrics

The aim of this work is to provide insights into multiple metrics clinical validation of deformable image registration and contour propagation methods in 4D lung radiotherapy planning. The following indices were analyzed and compared: Volume Difference (VD), Dice Similarity Coefficient (DSC), Positive Predictive Value (PPV) and Surface Distances (SD). The analysis was performed on three patient datasets, using as reference a ground-truth volume generated by means of Simultaneous Truth And Performance Level Estimation (STAPLE) algorithm from the outlines of five experts. Significant discrepancies in the quality assessment provided by the different metrics in all the examined cases were found. Metrics sensitivity was more evident in presence of image artifacts and particularly for tubular anatomical structures, such as esophagus or spinal cord. Volume Differences did not account for position and DSC exhibited criticalities due to its intrinsic symmetry ( i.e. over- and under-estimation of the reference contours cannot be discriminated) and dependency on the total volume of the structure. PPV analysis showed more robust performance, as each voxel concurs to the classification of the propagation, but was not able to detect inclusion of propagated and ground-truth volumes. Mesh distances could interpret the actual shape of the structures, but might report higher mismatches in case of large local differences in the contour surfaces. According to our study, the combination of VD and SD for the validation of contour propagation algorithms in 4D could provide the necessary failure detection accuracy.

Evaluating and modeling of photon beam attenuation by a standard treatment couch

The purpose of this study was to evaluate beam attenuation by treatment couch and build a treatment couch model in TPS to check for beam–couch intersection at the planning stage and deal with beam attenuation by treatment couch in dose calculation. In this study, a standard treatment couch, Siemens ZXT couch, has been incorporated into Pinnacle3 8.0 TPS, based on an existing TPS tool, model‐based segmentation (MBS). This was done by generating the couch's model from contours of the couch, together with the density information. Both the geometric and dosimetric accuracy of the couch model were evaluated. The test of beam–couch intersection prediction showed good agreement between predicted and measured results, and the differences were within 1° gantry rotation. For individual posterior oblique beams, the attenuation by metallic frames and PMMA couch top could reach nearly as high as 60% and 10%, respectively. For several posterior oblique beams (180°, 220°, 235°) that attenuated by the PMMA couch top, the calculated and measured dose distributions were compared. The dose differences at central axis were within 1%, and almost all points agreed with the calculations when the DD and DTA criteria of 3%/3 mm were adopted. The difference between calculated and measured attenuation factors were within 0.5%. This study demonstrates that the couch model created by MBS, which contains geometric and density information of the couch, can be used to detect the beam–couch intersection, and also is able to provide an accurate representation of the couch top attenuation properties in patient dose calculation.

PACS numbers: 87.55.D‐, 87.55.Gh, 87.55.km

Improving target delineation on 4-dimensional CT scans in stage I NSCLC using a deformable registration tool

INTRODUCTION

Correct target definition is crucial in stereotactic radiotherapy for lung tumors. We evaluated use of deformable registration (DR) for target contouring on 4-dimensional (4D) CT scans.

MATERIALS AND METHODS

Three clinicians contoured gross tumor volume (GTV) in an end-inspiration phase of 4DCT of 6 patients on two occasions. Two clinicians contoured GTVs in all phases of 4DCT and on maximum intensity projections (MIP). The initial GTV was auto-propagated to 9 other phases using a B-spline algorithm (VelocityAI). Internal target volumes (ITVs) generated were (i) ITV(10manual) encompassing all physician-contoured GTVs, (ii) ITV-MIP(optimized) from MIP after review of individual 4DCT phases, (iii) ITV(10deformed) encompassing auto-propagated GTVs using DR, and (iv) ITV(10deformed-optimized), from an ITV(10deformed) target that was modified to form a 'clinically optimal' ITV. Volume-overlaps were scored using Dice's Similarity Coefficients (DSCs).

RESULTS

Intra-clinician GTV reproducibility was greater than inter-clinician reproducibility (mean DSC 0.93 vs. 0.88, p<0.0004). In five of 6 patients, ITV-MIP(optimized) differed from the ITV(10deformed-optimized). In all patients, the DSC between ITV(10deformed-optimized) and ITV(10deformed) was higher than that between ITV(10deformed-optimized) and ITV-MIP(optimized) (p<0.02 T-test).

CONCLUSION

ITVs created in stage I tumors using DR were closer to 'clinically optimal' ITVs than was the case with a MIP-modified approach.

Accurate accumulation of dose for improved understanding of radiation effects in normal tissue

The actual distribution of radiation dose accumulated in normal tissues over the complete course of radiation therapy is, in general, poorly quantified. Differences in the patient anatomy between planning and treatment can occur gradually (e.g., tumor regression, resolution of edema) or relatively rapidly (e.g., bladder filling, breathing motion) and these undermine the accuracy of the planned dose distribution. Current efforts to maximize the therapeutic ratio require models that relate the true accumulated dose to clinical outcome. The needed accuracy can only be achieved through the development of robust methods that track the accumulation of dose within the various tissues in the body. Specific needs include the development of segmentation methods, tissue-mapping algorithms, uncertainty estimation, optimal schedules for image-based monitoring, and the development of informatics tools to support subsequent analysis. These developments will not only improve radiation outcomes modeling but will address the technical demands of the adaptive radiotherapy paradigm. The next 5 years need to see academia and industry bring these tools into the hands of the clinician and the clinical scientist.

How many sets of 4DCT images are sufficient to determine internal target volume for liver radiotherapy?

BACKGROUND AND PURPOSE

To determine the feasibility of using limited four-dimensional computed tomography (4DCT) images for treatment planning.

MATERIALS AND METHODS

The 4DCT scans of 16 patients with hepatocellular carcinoma (HCC) were analyzed. Gross tumor volumes (GTVs) were manually contoured on all 10 respiratory phases, and different internal clinical target volumes (ICTVs) were derived by encompassing volumes of the respective CTVs. Volume, position, and shape of ICTVs were calculated and compared.

RESULTS

The ICTV(2 phases), ICTV(3 phases), ICTV(4 phases), and ICTV(6 phases) all showed excellent agreement with ICTV(10 phases), and the ICTV(2 phases) encompassed ICTV(10 phases) by 94.1+/-1.8% on average. The 3D shift between the centers of mass of the ICTVs was only 0.6mm. The surface distance between ICTV(10 phases) and ICTV(2 phases) was 1.7+/-0.8mm in the left-right (LR) and anteroposterior (AP) directions.

CONCLUSIONS

Contouring two extreme phases at end-inhalation and end-exhalation is a reasonably safe and labor-saving method of deriving ITV for liver radiotherapy with low and medium tumor motion amplitude (1.6 cm). Whether the larger tumor movement affects the results is the subject of ongoing research.

Potential dosimetric benefits of four-dimensional radiation treatment planning

PURPOSE

To determine the extent of dosimetric differences between conventional three-dimensional (3D) dose calculations and four-dimensional (4D) dose calculations based on deformation of organ models.

METHODS AND MATERIALS

Four-dimensional dose calculations were retrospectively performed on computed tomography data sets for 15 patients with Stage III non-small-cell lung cancer, using a model-based deformable registration algorithm on a research version of a commercial radiation treatment planning system. Target volume coverage and doses to critical structures calculated using the 4D methodology were compared with those calculated using conventional 3D methodology.

RESULTS

For 11 of 15 patients, clinical target volume coverage was comparable in the 3D and 4D calculations, whereas for 7 of 15 patients, planning target volume coverage was comparable. For the other patients, the 4D calculation indicated a difference in target volume dose sufficiently great to warrant replanning. No correlations could be established between differences in 3D and 4D calculations and gross tumor volume size or extent of motion. Negligible differences were observed between 3D and 4D dose-volume relationships for normal anatomic structures.

CONCLUSIONS

Use of 4D dose calculations, when possible, helps ensure that target volumes will not be underirradiated when respiratory motion may affect the dose distribution.