Reconstruction of 3D lung models from 2D planning data sets for Hodgkin's lymphoma patients using combined deformable image registration and navigator channels

Machine learning for the prediction of pseudorealistic pediatric abdominal phantoms for radiation dose reconstruction

Purpose: Current phantoms used for the dose reconstruction of long-term childhood cancer survivors lack individualization. We design a method to predict highly individualized abdominal three-dimensional (3-D) phantoms automatically. Approach: We train machine learning (ML) models to map (2-D) patient features to 3-D organ-at-risk (OAR) metrics upon a database of 60 pediatric abdominal computed tomographies with liver and spleen segmentations. Next, we use the models in an automatic pipeline that outputs a personalized phantom given the patient's features, by assembling 3-D imaging from the database. A step to improve phantom realism (i.e., avoid OAR overlap) is included. We compare five ML algorithms, in terms of predicting OAR left-right (LR), anterior-posterior (AP), inferior-superior (IS) positions, and surface Dice-Sørensen coefficient (sDSC). Furthermore, two existing human-designed phantom construction criteria and two additional control methods are investigated for comparison. Results: Different ML algorithms result in similar test mean absolute errors: ∼ 8    mm for liver LR, IS, and spleen AP, IS; ∼ 5    mm for liver AP and spleen LR; ∼ 80 % for abdomen sDSC; and ∼ 60 % to 65% for liver and spleen sDSC. One ML algorithm (GP-GOMEA) significantly performs the best for 6/9 metrics. The control methods and the human-designed criteria in particular perform generally worse, sometimes substantially ( + 5 - mm error for spleen IS, - 10 % sDSC for liver). The automatic step to improve realism generally results in limited metric accuracy loss, but fails in one case (out of 60). Conclusion: Our ML-based pipeline leads to phantoms that are significantly and substantially more individualized than currently used human-designed criteria.

Abdominal organ position variation in children during image-guided radiotherapy

Background

Interfractional organ position variation might differ for abdominal organs and this could have consequences for defining safety margins. Therefore, the purpose of this study is to quantify interfractional position variations of abdominal organs in children in order to investigate possible correlations between abdominal organs and determine whether position variation is location-dependent.

Methods

For 20 children (2.2–17.8 years), we retrospectively analyzed 113 CBCTs acquired during the treatment course, which were registered to the reference CT to assess interfractional position variation of the liver, spleen, kidneys, and both diaphragm domes. Organ position variation was assessed in three orthogonal directions and relative to the bony anatomy. Diaphragm dome position variation was assessed in the cranial-caudal (CC) direction only. We investigated possible correlations between position variations of the organs (Spearman’s correlation test, ρ), and tested if organ position variations in the CC direction are related to the diaphragm dome position variations (linear regression analysis, R²) (both tests: significance level p < 0.05). Differences of variations of systematic (∑) and random errors (σ) between organs were tested (Bonferroni significance level p < 0.004).

Results

In all directions, correlations between liver and spleen position variations, and between right and left kidney position variations were weak (ρ ≤ 0.43). In the CC direction, the position variations of the right and left diaphragm domes were significantly, and stronger, correlated with position variations of the liver (R² = 0.55) and spleen (R² = 0.63), respectively, compared to the right (R² = 0.00) and left kidney (R² = 0.25). Differences in ∑ and σ between all organs were small and insignificant.

Conclusions

No (strong) correlations between interfractional position variations of abdominal organs in children were observed. From present results, we concluded that diaphragm dome position variations could be more representative for superiorly located abdominal (liver, spleen) organ position variations than for inferiorly located (kidneys) organ position variations. Differences of systematic and random errors between abdominal organs were small, suggesting that for margin definitions, there was insufficient evidence of a dependence of organ position variation on anatomical location.

Electronic supplementary material

The online version of this article (10.1186/s13014-018-1108-9) contains supplementary material, which is available to authorized users.

On the feasibility of automatically selecting similar patients in highly individualized radiotherapy dose reconstruction for historic data of pediatric cancer survivors

PURPOSE

The aim of this study is to establish the first step toward a novel and highly individualized three-dimensional (3D) dose distribution reconstruction method, based on CT scans and organ delineations of recently treated patients. Specifically, the feasibility of automatically selecting the CT scan of a recently treated childhood cancer patient who is similar to a given historically treated child who suffered from Wilms' tumor is assessed.

METHODS

A cohort of 37 recently treated children between 2- and 6-yr old are considered. Five potential notions of ground-truth similarity are proposed, each focusing on different anatomical aspects. These notions are automatically computed from CT scans of the abdomen and 3D organ delineations (liver, spleen, spinal cord, external body contour). The first is based on deformable image registration, the second on the Dice similarity coefficient, the third on the Hausdorff distance, the fourth on pairwise organ distances, and the last is computed by means of the overlap volume histogram. The relationship between typically available features of historically treated patients and the proposed ground-truth notions of similarity is studied by adopting state-of-the-art machine learning techniques, including random forest. Also, the feasibility of automatically selecting the most similar patient is assessed by comparing ground-truth rankings of similarity with predicted rankings.

RESULTS

Similarities (mainly) based on the external abdomen shape and on the pairwise organ distances are highly correlated (Pearson r_p ≥ 0.70) and are successfully modeled with random forests based on historically recorded features (pseudo-R² ≥ 0.69). In contrast, similarities based on the shape of internal organs cannot be modeled. For the similarities that random forest can reliably model, an estimation of feature relevance indicates that abdominal diameters and weight are the most important. Experiments on automatically selecting similar patients lead to coarse, yet quite robust results: the most similar patient is retrieved only 22% of the times, however, the error in worst-case scenarios is limited, with the fourth most similar patient being retrieved.

CONCLUSIONS

Results demonstrate that automatically selecting similar patients is feasible when focusing on the shape of the external abdomen and on the position of internal organs. Moreover, whereas the common practice in phantom-based dose reconstruction is to select a representative phantom using age, height, and weight as discriminant factors for any treatment scenario, our analysis on abdominal tumor treatment for children shows that the most relevant features are weight and the anterior-posterior and left-right abdominal diameters.

Late Cardiac Toxicity After Mediastinal Radiation Therapy for Hodgkin Lymphoma: Contributions of Coronary Artery and Whole Heart Dose-Volume Variables to Risk Prediction

PURPOSE

Mediastinal radiation therapy (RT) for Hodgkin lymphoma (HL) is associated with late cardiotoxicity, but there are limited data to indicate which dosimetric parameters are most valuable for predicting this risk. This study investigated which whole heart dosimetric measurements provide the most information regarding late cardiotoxicity, and whether coronary artery dosimetry was more predictive of this outcome than whole heart dosimetry.

METHODS AND MATERIALS

A random sample of 125 HL patients treated with mediastinal RT was selected, and 3-dimensional cardiac dose-volume data were generated from historical plans using validated methods. Cardiac events were determined by linking patients to population-based datasets of inpatient and same-day hospitalizations and same-day procedures. Variables collected for the whole heart and 3 coronary arteries included the following: Dmean, Dmax, Dmin, dose homogeneity, V5, V10, V20, and V30. Multivariable competing risk regression models were generated for the whole heart and coronary arteries.

RESULTS

There were 44 cardiac events documented, of which 70% were ischemic. The best multivariable model included the following covariates: whole heart Dmean (hazard ratio [HR] 1.09, P=.0083), dose homogeneity (HR 0.94, P=.0034), male sex (HR 2.31, P=.014), and age (HR 1.03, P=.0049). When any adverse cardiac event was the outcome, models using coronary artery variables did not perform better than models using whole heart variables. However, in a subanalysis of ischemic cardiac events only, the model using coronary artery variables was superior to the whole heart model and included the following covariates: age (HR 1.05, P<.001), volume of left anterior descending artery receiving 5 Gy (HR 0.98, P=.003), and volume of left circumflex artery receiving 20 Gy (HR 1.03, P<.001).

CONCLUSION

In addition to higher mean heart dose, increasing inhomogeneity in cardiac dose was associated with a greater risk of late cardiac effects. When all types of cardiotoxicity were evaluated, the whole heart variable model outperformed the coronary artery models. However, when events were limited to ischemic cardiotoxicity, the coronary artery-based model was superior.

A Biomechanical Modeling Guided CBCT Estimation Technique

Two-dimensional-to-three-dimensional (2D-3D) deformation has emerged as a new technique to estimate cone-beam computed tomography (CBCT) images. The technique is based on deforming a prior high-quality 3D CT/CBCT image to form a new CBCT image, guided by limited-view 2D projections. The accuracy of this intensity-based technique, however, is often limited in low-contrast image regions with subtle intensity differences. The solved deformation vector fields (DVFs) can also be biomechanically unrealistic. To address these problems, we have developed a biomechanical modeling guided CBCT estimation technique (Bio-CBCT-est) by combining 2D-3D deformation with finite element analysis (FEA)-based biomechanical modeling of anatomical structures. Specifically, Bio-CBCT-est first extracts the 2D-3D deformation-generated displacement vectors at the high-contrast anatomical structure boundaries. The extracted surface deformation fields are subsequently used as the boundary conditions to drive structure-based FEA to correct and fine-tune the overall deformation fields, especially those at low-contrast regions within the structure. The resulting FEA-corrected deformation fields are then fed back into 2D-3D deformation to form an iterative loop, combining the benefits of intensity-based deformation and biomechanical modeling for CBCT estimation. Using eleven lung cancer patient cases, the accuracy of the Bio-CBCT-est technique has been compared to that of the 2D-3D deformation technique and the traditional CBCT reconstruction techniques. The accuracy was evaluated in the image domain, and also in the DVF domain through clinician-tracked lung landmarks.

Image-guided radiation therapy in lymphoma management

Image-guided radiation therapy (IGRT) is a process of incorporating imaging techniques such as computed tomography (CT), magnetic resonance imaging (MRI), Positron emission tomography (PET), and ultrasound (US) during radiation therapy (RT) to improve treatment accuracy. It allows real-time or near real-time visualization of anatomical information to ensure that the target is in its position as planned. In addition, changes in tumor volume and location due to organ motion during treatment can be also compensated. IGRT has been gaining popularity and acceptance rapidly in RT over the past 10 years, and many published data have been reported on prostate, bladder, head and neck, and gastrointestinal cancers. However, the role of IGRT in lymphoma management is not well defined as there are only very limited published data currently available. The scope of this paper is to review the current use of IGRT in the management of lymphoma. The technical and clinical aspects of IGRT, lymphoma imaging studies, the current role of IGRT in lymphoma management and future directions will be discussed.

Individualized 3D reconstruction of normal tissue dose for patients with long-term follow-up: a step toward understanding dose risk for late toxicity

PURPOSE

Understanding the relationship between normal tissue dose and delayed radiation toxicity is an important component of developing more effective radiation therapy. Late outcome data are generally available only for patients who have undergone 2-dimensional (2D) treatment plans. The purpose of this study was to evaluate the accuracy of 3D normal tissue dosimetry derived from reconstructed 2D treatment plans in Hodgkin's lymphoma (HL) patients.

METHODS AND MATERIALS

Three-dimensional lung, heart, and breast volumes were reconstructed from 2D planning radiographs for HL patients who received mediastinal radiation therapy. For each organ, a reference 3D organ was modified with patient-specific structural information, using deformable image processing software. Radiation therapy plans were reconstructed by applying treatment parameters obtained from patient records to the reconstructed 3D volumes. For each reconstructed organ mean dose (Dmean) and volumes covered by at least 5 Gy (V5) and 20 Gy (V20) were calculated. This process was performed for 15 patients who had both 2D and 3D planning data available to compare the reconstructed normal tissue doses with those derived from the primary CT planning data and also for 10 historically treated patients with only 2D imaging available.

RESULTS

For patients with 3D planning data, the normal tissue doses could be reconstructed accurately using 2D planning data. Median differences in Dmean between reconstructed and actual plans were 0.18 Gy (lungs), -0.15 Gy (heart), and 0.30 Gy (breasts). Median difference in V5 and V20 were less than 2% for each organ. Reconstructed 3D dosimetry was substantially higher in historical mantle-field treatments than contemporary involved-field mediastinal treatments: average Dmean values were 15.2 Gy vs 10.6 Gy (lungs), 27.0 Gy vs 14.3 Gy (heart), and 8.0 Gy vs 3.2 Gy (breasts).

CONCLUSIONS

Three-dimensional reconstruction of absorbed dose to organs at risk can be estimated accurately many years after exposure by using limited 2D data. Compared to contemporary involved-field treatments, normal tissue doses were significantly higher in historical mantle-field treatments. These methods build capacity to quantify the relationship between 3D normal tissue dose and observed late effects.

Methods to assess adverse health-related outcomes in cancer survivors

Designing a study focused on adverse health-related outcomes among cancer survivors is complex. Similarly, reading and interpreting the findings of a survivorship-focused study requires an appreciation of the complexities of study design, potential biases, confounding factors, and other limitations. The topic areas are broad--study design, comparison populations, measures of risk, key health outcomes of interest, potential modifying factors to consider. With brevity, this article includes basic information to consider within these areas as well as examples and concepts intended to advance the science of survivorship research and encourage further reading and exploration.

Navigator channel adaptation to reconstruct three dimensional heart volumes from two dimensional radiotherapy planning data

Background

Biologically-based models that utilize 3D radiation dosimetry data to estimate the risk of late cardiac effects could have significant utility for planning radiotherapy in young patients. A major challenge arises from having only 2D treatment planning data for patients with long-term follow-up. In this study, we evaluate the accuracy of an advanced deformable image registration (DIR) and navigator channels (NC) adaptation technique to reconstruct 3D heart volumes from 2D radiotherapy planning images for Hodgkin's Lymphoma (HL) patients.

Methods

Planning CT images were obtained for 50 HL patients who underwent mediastinal radiotherapy. Twelve image sets (6 male, 6 female) were used to construct a male and a female population heart model, which was registered to 23 HL "Reference" patients' CT images using a DIR algorithm, MORFEUS. This generated a series of population-to-Reference patient specific 3D deformation maps. The technique was independently tested on 15 additional "Test" patients by reconstructing their 3D heart volumes using 2D digitally reconstructed radiographs (DRR). The technique involved: 1) identifying a matching Reference patient for each Test patient using thorax measurements, 2) placement of six NCs on matching Reference and Test patients' DRRs to capture differences in significant heart curvatures, 3) adapting the population-to-Reference patient-specific deformation maps to generate population-to-Test patient-specific deformation maps using linear and bilinear interpolation methods, 4) applying population-to-Test patient specific deformation to the population model to reconstruct Test-patient specific 3D heart models. The percentage volume overlap between the NC-adapted reconstruction and actual Test patient's true heart volume was calculated using the Dice coefficient.

Results

The average Dice coefficient expressed as a percentage between the NC-adapted and actual Test model was 89.4 ± 2.8%. The modified NC adaptation technique made significant improvements to the population deformation heart models (p = 0.01). As standard evaluation, the residual Dice error after adaptation was comparable to the volumetric differences observed in free-breathing heart volumes (p = 0.62).

Conclusions

The reconstruction technique described generates accurate 3D heart models from limited 2D planning data. This development could potentially be used to retrospectively calculate delivered dose to the heart for historically treated patients and thereby provide a better understanding of late radiation-related cardiac effects.