3D Lung Volume Research Articles

Deep-Learning Model Prediction of Radiation Pneumonitis Using Pretreatment Chest Computed Tomography and Clinical Factors.

Objectives: This study aimed to build a comprehensive deep-learning model for the prediction of radiation pneumonitis using chest computed tomography (CT), clinical, dosimetric, and laboratory data. Introduction: Radiation therapy is an effective tool for treating patients with lung cancer. Despite its effectiveness, the risk of radiation pneumonitis limits its application. Although several studies have demonstrated models to predict radiation pneumonitis, no reliable model has been developed yet. Herein, we developed prediction models using pretreatment chest CT and various clinical data to assess the likelihood of radiation pneumonitis in lung cancer patients. Methods: This retrospective study analyzed 3-dimensional (3D) lung volume data from chest CT scans and 27 features including dosimetric, clinical, and laboratory data from 548 patients who were treated at our institution between 2010 and 2021. We developed a neural network, named MergeNet, which processes lung 3D CT, clinical, dosimetric, and laboratory data. The MergeNet integrates a convolutional neural network with subsequent fully connected layers. A support vector machine (SVM) and light gradient boosting machine (LGBM) model were also implemented for comparison. For comparison, the convolution-only neural network was implemented as well. Three-dimensional Resnet-10 network and 4-fold cross-validation were used. Results: Classification performance was quantified by using the area under the receiver operative characteristic curve (AUC) metrics. MergeNet showed the AUC of 0.689. SVM, LGBM, and convolution-only networks showed AUCs of 0.525, 0.541, and 0.550, respectively. Application of DeLong test to pairs of receiver operating characteristic curves respectively yielded P values of .001 for the MergeNet-SVM pair and 0.001 for the MergeNet-LGBM pair. Conclusion: The MergeNet model, which incorporates chest CT, clinical, dosimetric, and laboratory data, demonstrated superior performance compared to other models. However, since its prediction performance has not yet reached an efficient level for clinical application, further research is required. Contribution: This study showed that MergeNet may be an effective means to predict radiation pneumonitis. Various predictive factors can be used together for the radiation pneumonitis prediction task via the MergeNet.

Three-Dimensional Ultrasound Evaluation of Lung Volume in Fetuses with Abdominal Wall Defect

Introduction: Abdominal wall defects (AWDs) interfere with postnatal respiratory parameters. We aimed to evaluate lung volume (LV) in fetuses with AWD using three-dimensional (3D) ultrasound (US) and to correlate AWD with the type (omphalocele and gastroschisis) and size of the defect and neonatal morbidity and mortality. Methods: This prospective observational study included 72 pregnant women with fetuses with AWD and a gestational age <25 weeks. The data on abdominal volume, 3D US LV, and herniated volume were acquired every 4 weeks up to 33 weeks. LV was compared with normal reference curves and correlated with abdominal and herniated volumes. Results: Omphalocele (p < 0.001) and gastroschisis (p < 0.001) fetuses had smaller LV than normal fetuses. LV was positively correlated with abdominal volume (omphalocele, r = 0.86; gastroschisis, r = 0.88), whereas LV was negatively correlated with omphalocele-herniated volume/abdominal volume (p < 0.001, r = −0.51). LV was smaller in omphalocele fetuses that died (p = 0.002), were intubated (p = 0.02), or had secondary closure (p < 0.001). In gastroschisis, a smaller LV was observed in fetuses discharged using oxygen (p = 0.002). Conclusion: Fetuses with AWD had smaller 3D LV than normal fetuses. Fetal abdominal volume was inversely correlated with LV. In omphalocele fetuses, a smaller LV was associated with neonatal mortality and morbidity.

Surface Projection of CT Reconstructed Lung Volume for Regional Lung Mechanics Estimation: A Preliminary Analysis

Estimating regional lung volume along bedside is critical for creating a fine level of digital twins model for optimized setting of mechanical ventilation. Non-contact techniques with structured light is capable of reconstructing chest motion to incorporate with ventilator recorded volume to assess the regional lung volume changes between left and right lung, while its extension to a higher resolution of estimation requires a relation map between the surface motion and inside lung capacity distribution. To establish the relation map, a 2D projection of 3D lung CT lung volume is proposed to represent the lung capacity distribution on the chest surface plane in this study. A convolutional neural network (CNN) is designed to implement lung segmentation and reconstruct 3D lung volume via CT scans. The 2D projection map is then compared to the recorded forced vital capacity (FVC) as a surrogate of tidal volume. The test was conducted against two CT datasets for training CNN model and build the 2D surface mapping. The results show a good correlation of R squared value of 0.71 between the overall lung capacity via the proposed 2D projection map against the FVC, thus showing the potential of back calculate the regional volume distribution using measured ventilator tidal volume and structured light surface motion.

The Effect of Disease Type on Changes in Total Lung Volume after Lung Transplantation Measured by Three-Dimensional (3D) CT Reconstruction

<h3>Purpose</h3> It is accepted that a donor's predicted total lung capacity (pTLC) can be 75-125% of the recipient's pTLC in lung transplantation; however, ‘optimal graft size' is not clearly defined, especially in restricted lung disease (RD) patients. These disagreements exist as it is difficult to accurately measure lung size before and after transplantation and it is also difficult to predict and evaluate the extent of re-expansion of the restricted thoracic cavity after transplantation. This study examined changes in lung volume after lung transplantation using a 3D lung volume measurement from CT scans before and after surgery under the assumption that the patient's thoracic cavity would return to the size of the patient without underlying lung disease after surgery. <h3>Methods</h3> A total of 113 patients, who underwent lung transplantation between September 2012 and April 2019 in our institute, and met the inclusion criteria were enrolled in the study. This included 88 patients who were diagnosed with RD and 25 patients who were diagnosed with non-RD. Patients were further divided into oversized and undersized groups according to the donor and recipient pTLC ratio, and a total of 4 groups were studied. Preoperative and postoperative lung volume on CT scans were measured using the synapse 3D program (Fuji film, Japan). <h3>Results</h3> RD patients showed a significant increase in total lung volume (TLV) after surgery, while non-RD patients showed a decrease in TLV. These changes showed the greatest increase in the RD group within 1 year and continuously increased up to 5 years. In the non-RD group, TLV decreased within 3 months of surgery. These changes were not related to graft size. <h3>Conclusion</h3> In lung transplant patients, the thoracic cavity re-expanded in RD patients and reduced in the non-RD patients, returning to healthy values without lung disease status in terms of TLV.

Histogram based analysis of lung perfusion of children after congenital diaphragmatic hernia repair

ObjectivesTo investigate a histogram based approach to characterize the distribution of perfusion in the whole left and right lung by descriptive statistics and to show how histograms could be used to visually explore perfusion defects in two year old children after Congenital Diaphragmatic Hernia (CDH) repair. Methods28 children (age of 24.2±1.7months; all left sided hernia; 9 after extracorporeal membrane oxygenation therapy) underwent quantitative DCE-MRI of the lung. Segmentations of left and right lung were manually drawn to mask the calculated pulmonary blood flow maps and then to derive histograms for each lung side. Individual and group wise analysis of histograms of left and right lung was performed. ResultsIpsilateral and contralateral lung show significant difference in shape and descriptive statistics derived from the histogram (Wilcoxon signed-rank test, p<0.05) on group wise and individual level. Subgroup analysis (patients with vs without ECMO therapy) showed no significant differences using histogram derived parameters. ConclusionsHistogram analysis can be a valuable tool to characterize and visualize whole lung perfusion of children after CDH repair. It allows for several possibilities to analyze the data, either describing the perfusion differences between the right and left lung but also to explore and visualize localized perfusion patterns in the 3D lung volume. Subgroup analysis will be possible given sufficient sample sizes.

HOSVD-Based 3D Active Appearance Model: Segmentation of Lung Fields in CT Images.

An Active Appearance Model (AAM) is a computer vision model which can be used to effectively segment lung fields in CT images. However, the fitting result is often inadequate when the lungs are affected by high-density pathologies. To overcome this problem, we propose a Higher-order Singular Value Decomposition (HOSVD)-based Three-dimensional (3D) AAM. An evaluation was performed on 310 diseased lungs form the Lung Image Database Consortium Image Collection. Other contemporary AAMs operate directly on patterns represented by vectors, i.e., before applying the AAM to a 3D lung volume,it has to be vectorized first into a vector pattern by some technique like concatenation. However, some implicit structural or local contextual information may be lost in this transformation. According to the nature of the 3D lung volume, HOSVD is introduced to represent and process the lung in tensor space. Our method can not only directly operate on the original 3D tensor patterns, but also efficiently reduce the computer memory usage. The evaluation resulted in an average Dice coefficient of 97.0% ± 0.59%, a mean absolute surface distance error of 1.0403 ± 0.5716mm, a mean border positioning errors of 0.9187 ± 0.5381pixel, and a Hausdorff Distance of 20.4064 ± 4.3855, respectively. Experimental results showed that our methods delivered significant and better segmentation results, compared with the three other model-based lung segmentation approaches, namely 3D Snake, 3D ASM and 3D AAM.

SU‐E‐J‐238: First‐Order Approximation of Time‐Resolved 4DMRI From Cine 2DMRI and Respiratory‐Correlated 4DMRI

Purpose:Cine 2DMRI is useful in MR‐guided radiotherapy but it lacks volumetric information. We explore the feasibility of estimating timeresolved (TR) 4DMRI based on cine 2DMRI and respiratory‐correlated (RC) 4DMRI though simulation.Methods:We hypothesize that a volumetric image during free breathing can be approximated by interpolation among 3DMRI image sets generated from a RC‐4DMRI. Two patients’ RC‐4DMRI with 4 or 5 phases were used to generate additional 3DMRI by interpolation. For each patient, six libraries were created to have total 5‐to‐35 3DMRI images by 0–6 equi‐spaced tri‐linear interpolation between adjacent and full‐inhalation/full‐exhalation phases. Sagittal cine 2DMRI were generated from reference 3DMRIs created from separate, unique interpolations from the original RC‐4DMRI. To test if accurate 3DMRI could be generated through rigid registration of the cine 2DMRI to the 3DMRI libraries, each sagittal 2DMRI was registered to sagittal cuts in the same location in the 3DMRI within each library to identify the two best matches: one with greater lung volume and one with smaller. A final interpolation between the corresponding 3DMRI was then performed to produce the first‐order‐approximation (FOA) 3DMRI. The quality and performance of the FOA as a function of library size was assessed using both the difference in lung volume and average voxel intensity between the FOA and the reference 3DMRI.Results:The discrepancy between the FOA and reference 3DMRI decreases as the library size increases. The 3D lung volume difference decreases from 5–15% to 1–2% as the library size increases from 5 to 35 image sets. The average difference in lung voxel intensity decreases from 7–8 to 5–6 with the lung intensity being 0–135.Conclusion:This study indicates that the quality of FOA 3DMRI improves with increasing 3DMRI library size. On‐going investigations will test this approach using actual cine 2DMRI and introduce a higher order approximation for improvements.This study is in part supported by NIH (U54CA137788 and U54CA132378).

Prediction of lethal pulmonary hypoplasia among high-risk fetuses via 2D and 3D ultrasonography

ObjectiveTo compare accuracy between 3D and 2D ultrasonography for predicting lethal pulmonary hypoplasia (LPH) among high-risk fetuses. MethodsIn a cross-sectional prospective study at a fetal medicine referral center in Brazil, bilateral fetal lung scans were evaluated for 54 fetuses with suspected LPH between May 2008 and June 2011. Measurements for predicting LPH were ultrasonographic fetal lung volume/estimated fetal weight ratio (US-FLW), observed/expected thoracic circumference (o/e-TC), observed/expected thoracic circumference/abdominal circumference (o/e-TC/AC), observed/expected thoracic area/heart area (o/e-TA/HA), observed/expected amniotic fluid index (o/e-AFI), and observed/expected total fetal lung volume (o/e-Tot-FLV). To evaluate accuracy in predicting LPH, receiver operating characteristic (ROC) curves were calculated, and areas under the curves (AUCs) were compared. The intraclass correlation coefficient (ICC) was used to assess 3D lung volume reproducibility. ResultsData were compared for 47 newborns, of whom 34 had LPH (perinatal mortality 74.5%). The AUCs for predicting LPH were 0.93, 0.69, 0.55, 0.46, 0.71, and 0.86 for US-FLW, o/e-TC, o/e-TC/AC, o/e-TA/HA, o/e–AFI, and o/e-Tot–FLV, respectively. No significant intraobserver difference was observed in measurements of the right (ICC, 0.973; P<0.0001) or left (ICC, 0.950; P<0.0001) lung volumes. ConclusionAmong high-risk fetuses, US-FLW by 2D/3D and o/e-Tot-FLV by 3D ultrasonography were more accurate than 2D parameters for predicting LPH.

Prediction of pulmonary hypoplasia in mid‐trimester preterm prelabor rupture of membranes: research or clinical practice?

Prediction of pulmonary hypoplasia in mid‐trimester preterm prelabor rupture of membranes: research or clinical practice?

Two‐ or three‐dimensional ultrasonography: which is the best predictor of pulmonary hypoplasia?

The aim of our study is to compare the three-dimensional (3D) ultrasound fetal lung volume measurements with two-dimensional (2D) ultrasound biometric parameters in predicting pulmonary hypoplasia (PH). From June 2006 to November 2008, we identified 35 fetuses at high risk for developing PH, including premature preterm rupture of membranes (17), fetal skeletal malformations (7), hydrothorax (7), and bilateral renal dysplasia (4). 3D lung volumes adjusted for gestational age (GA) or estimated fetal weight (EFW), thoracic circumference adjusted for GA or femur length, thoracic/abdominal circumference ratio, and thoracic/heart area ratio were measured. Three infants were excluded because no clinical or histological information regarding lung hypoplasia was available. Of the 32 remaining infants, 13 (41%) were diagnosed with PH at postmortem examination or by clinical and radiological examination. 3D lung volume measurements had a better diagnostic accuracy for predicting PH either when adjusted for GA [sensitivity, 12/13 (92%); specificity, 16/19 (84%); positive predictive value 12/15 (80%); negative predictive value, 16/17(94%)] or EFW [sensitivity, 11/13 (85%); specificity, 18/19 (94%); positive predictive value 11/12(92%); negative predictive value, 18/20(90%)] compared to the 2D biometric measurement. 3D lung volume measurements seem to be useful in the prenatal prediction of PH.

SU‐FF‐I‐155: A Novel Technique to Generate 3D Lung Volumes From 2D Hodgkin Lymphoma Planning Datasets Using Combined Deformable Image Registration and Navigator Channels

Purpose: To validate the use of deformable image registration and navigator channels (NC) to generate 3D lung volumes from 2D images, for retrospective dose calculation of Hodgkin Lymphoma (HL) patients treated with 2D planning. Methods and Materials: Forty‐seven HL patients with 3D images were acquired. Six patients were used to construct a population lung model, which was then registered, using a biomechanical model‐based deformable registration algorithm, MORFEUS, into 24 additional HL ‘reference’ patients to develop a lung motion model. This model was refined to describe the 3D lung volume and position of 17 HL ‘test’ patients, using only information in the 2D DRR constructed from 3D images. The refinement was performed by: 1) matching a reference patient to each test patient (using lung width and length measurements), 2) placing four NCs, or small regions of interest, on the lung boundary of the reference patient and automatically placing the corresponding NC on the test image in the same geometric position, 3) converting the image intensity in the NC into a 1D shift to match that location, 4) refining the lung motion model using the four 1D shifts to describe the 3D volume and position of the test patient's lung. The refined lung model was then compared to the actual 3D lung contour of the 17 test patients by computing the volume overlap. Results: The average percentage overlapping and non‐overlapping volumes between the NC‐refined lung model and test lungs were 89.2±3.9% (Right Lung=88.8%; Left Lung=89.6%) and 10.8±3.9% (Right=11.2%; Left=10.4%), respectively. NCs made a statistically significant improvement to the population lung model shape (T‐test: p &lt; 0.05). Conclusion: This technique will be used to generate 3D lung volumes for retrospective patients treated with 2D planning for longitudinal studies to improve understanding of the dose‐risk relationship and various late effects.

Congenital Diaphragmatic Hernia: 2D Lung Area and 3D Lung Volume Measurements of the Contralateral Lung to Predict Postnatal Outcome

Objectives: The aim of the study was to evaluate the prognostic utility of 2D lung area and 3D lung volume measurements of the contralateral lung in infants with congenital diaphragmatic hernia. Methods: At 18–37 weeks’ gestation between 1 and 5 scans of the contralateral fetal lung were obtained in 6 pregnancies complicated by congenital diaphragmatic hernia (5 left- and 1 right-sided). Lung volume measurements were compared with reference curves for gestational age and estimated fetal weight obtained from uncomplicated pregnancies. Lung area measurements were compared with reference nomograms. Results: Three infants survived and 3 died. Lung volume measurements versus gestational age were beneath the 5th percentile for the nonsurviving infants and within the normal ranges for the surviving infants. When comparing the observed/expected lung volume and lung area ratios of the first measurements with the ratios at the last visit before birth, the ratios of the infants who subsequently died decreased whereas the ratios of the infants who survived remained unchanged or increased. Conclusions: In the case of congenital diaphragmatic hernia, 3D lung volume measurements seem to be a good predictor of outcome but longitudinal measurement also provides important additional information. Larger studies are necessary to confirm these results.

Comparison of the two-dimensional and multiplanar methods and establishment of a new constant for the measurement of fetal lung volume

Objectives. To compare the two-dimensional (2D) and multiplanar methods in the measurement of lung volume in normal fetuses, to obtain a new constant to be incorporated into the 2D equation, and to apply the new equation in fetuses with pulmonary hypoplasia (PH) confirmed postnatally.Methods. A cross-sectional study was performed on 51 pregnant women at between 20 and 35 weeks of gestation. The ellipsoid formula (x × y × z × 0.52) was used to calculate lung volume by the 2D method. A sequence of multiple planes with 2.0-mm intervals was used with the multiplanar method. In order to compare the techniques, the intra-class correlation coefficient (ICC) and the Student's t-test (p) were used. First-order linear regressions were used to establish a new constant, with three-dimensional (3D) lung volume as dependent variable and gestational age and 2D volume as independent variables. In order to validate it, the new equation was applied to 11 fetuses with confirmed lethal PH.Results. The mean volumes obtained by the 2D method were overestimated when compared to the multiplanar method (right lung: 23.87 vs. 18.26 mL, respectively, p < 0.001 and left lung: 16.18 vs. 14.33 mL, respectively, p = 0.008). Using a first-order polynomial regression, new constants were obtained for the right lung (0.152) and for the left lung (0.167). When compared to the traditional formula, the new equation presented higher sensitivity (18.1%) in predicting lethal PH.Conclusion. The recalculated 2D equation can be a promising alternative to 3D ultrasonography in the prediction of PH.

Predicting pulmonary hypoplasia with 2- or 3-dimensional ultrasonography in complicated pregnancies

The aim of this study was to compare 3-dimensional (3D) lung volume measurements with 2-dimensional (2D) biometric parameters in predicting pulmonary hypoplasia in complicated pregnancies. In this prospective study, 1-4 scans of the fetal lungs were obtained in 33 pregnancies complicated by various disorders or complications with regard to pulmonary hypoplasia. The 3D lung volumes vs gestational age or estimated fetal weight, the thoracic circumference vs gestational age or femur length, the thoracic/abdominal circumference ratio, and the thoracic/heart area ratio were measured. Of the 33 infants, 16 (48.5%) were diagnosed with pulmonary hypoplasia on postmortem examination or the clinical and radiological presentation. Three dimensional lung volume measurements had a better diagnostic accuracy for predicting pulmonary hypoplasia (sensitivity, 94%; specificity, 82%; positive predictive value [PPV], 83%; negative predictive value [NPV], 93%), compared with the best 2D biometric measurement thoracic/heart area ratio (sensitivity, 94%; specificity, 47%; PPV, 63%; NPV, 89%). 3D lung volume measurements seem to be useful in predicting pulmonary hypoplasia prenatally.

Two‐ or three‐dimensional ultrasonography to predict pulmonary hypoplasia in pregnancies complicated by preterm premature rupture of the membranes

The purpose of this study was to compare 3D lung volume measurements with 2D biometric parameters in predicting pulmonary hypoplasia in pregnancies complicated by preterm premature rupture of the membranes (PPROM). In this prospective study, 18 pregnancies complicated by PPROMs at a mean 21 weeks' gestation (range 14-32 weeks) were examined. The 3D lung volume measurements and the following 2D biometric parameters were measured: thoracic circumference (TC) versus gestational age or femur length (FL), the TC/abdominal circumference (AC) ratio and the thoracic area/heart area (TA/HA) ratio. The sensitivity, specificity, positive and negative predictive value of each measurement to diagnose pulmonary hypoplasia were compared. Pulmonary hypoplasia was diagnosed on the basis of clinical, radiological and/or pathologic criteria. The incidence of pulmonary hypoplasia was 33.3%. The best diagnostic accuracy for predicting pulmonary hypoplasia was achieved using the 3D lung volume measurements versus gestational age (sensitivity 83%, specificity 100%, positive predictive value 100% and negative predictive value 92%). Three-dimensional lung volume measurements seem to be promising in predicting pulmonary hypoplasia prenatally in pregnancies complicated by PPROM.

Virtual Bronchoscopy

Patients potentially suffer and are exposed to danger during invasive bronchoscopic diagnosis and surgery. In order to reduce this hazardous risk, we have developed an interactive virtual environment for the simulation of bronchoscopy (in short, called "virtual bronchoscopy"). Because of this state-of-the-art application, medical doctors can now obtain pre-operative information and perform pilot examinations in a virtual environment without any invasive or needless surgery. This 3D lung volume data of the patient is first acquired from CT and/or MRI scanning, without any pain being inflicted upon the patient. Then a vessel-tracking process is used to extract the patient's bronchial tree from the data. It is important to note that while manual tracking is tedious and labor-intensive, fully automatic tracking may not be as reliable in such a critical medical application. Thus a semi-automatic tracking technique called the Intelligent Path Tracker, which provides automation and sufficient user control during the tracking process, is most useful. This methodology is applied to a virtual bronchoscopy session, where doctors can use a 3D pen input device to navigate and visualize the bronchial tree of patients in a natural and interactive manner. To support an interactive frame rate, we also propose a new volume rendering acceleration technique, named IsoRegion Leaping. Through this technique visualization is further accelerated using a distributed rendering process based upon a TCP/IP network of low-cost PCs. Combining these approaches enables a 256x256x256 volumetric data representation of a human lung to be navigated and visualized at a frame rate of over 10 Hz in our virtual bronchoscopy system.