Dosimetric Impact of Delineation and Motion Uncertainties on the Heart and Substructures in Lung Cancer Radiotherapy

AIMS

Delineation variations and organ motion produce difficult-to-quantify uncertainties in planned radiation doses to targets and organs at risk. Similar to manual contouring, most automatic segmentation tools generate single delineations per structure; however, this does not indicate the range of clinically acceptable delineations. This study develops a method to generate a range of automatic cardiac structure segmentations, incorporating motion and delineation uncertainty, and evaluates the dosimetric impact in lung cancer.

MATERIALS AND METHODS

Eighteen cardiac structures were delineated using a locally developed auto-segmentation tool. It was applied to lung cancer planning CTs for 27 curative (planned dose ≥50 Gy) cases, and delineation variations were estimated by using ten mapping-atlases to provide separate substructure segmentations. Motion-related cardiac segmentation variations were estimated by auto-contouring structures on ten respiratory phases for 9/27 cases that had 4D-planning CTs. Dose volume histograms (DVHs) incorporating these variations were generated for comparison.

RESULTS

Variations in mean doses (Dmean), defined as the range in values across ten feasible auto-segmentations, were calculated for each cardiac substructure. Over the study cohort the median variations for delineation uncertainty and motion were 2.20-11.09 Gy and 0.72-4.06 Gy, respectively. As relative values, variations in Dmean were between 18.7%-65.3% and 7.8%-32.5% for delineation uncertainty and motion, respectively. Doses vary depending on the individual planned dose distribution, not simply on segmentation differences, with larger dose variations to cardiac structures lying within areas of steep dose gradient.

CONCLUSION

Radiotherapy dose uncertainties from delineation variations and respiratory-related heart motion were quantified using a cardiac substructure automatic segmentation tool. This predicts the 'dose range' where doses to structures are most likely to fall, rather than single DVH curves. This enables consideration of these uncertainties in cardiotoxicity research and for future plan optimisation. The tool was designed for cardiac structures, but similar methods are potentially applicable to other OARs.

Validation of a Fully Automated Hybrid Deep Learning Cardiac Substructure Segmentation Tool for Contouring and Dose Evaluation in Lung Cancer Radiotherapy

BACKGROUND AND PURPOSE

Accurate and consistent delineation of cardiac substructures is challenging. The aim of this work was to validate a novel segmentation tool for automatic delineation of cardiac structures and subsequent dose evaluation, with potential application in clinical settings and large-scale radiation-related cardiotoxicity studies.

MATERIALS AND METHODS

A recently developed hybrid method for automatic segmentation of 18 cardiac structures, combining deep learning, multi-atlas mapping and geometric segmentation of small challenging substructures, was independently validated on 30 lung cancer cases. These included anatomical and imaging variations, such as tumour abutting heart, lung collapse and metal artefacts. Automatic segmentations were compared with manual contours of the 18 structures using quantitative metrics, including Dice similarity coefficient (DSC), mean distance to agreement (MDA) and dose comparisons.

RESULTS

A comparison of manual and automatic contours across all cases showed a median DSC of 0.75-0.93 and a median MDA of 2.09-3.34 mm for whole heart and chambers. The median MDA for great vessels, coronary arteries, cardiac valves, sinoatrial and atrioventricular conduction nodes was 3.01-8.54 mm. For the 27 cases treated with curative intent (planned target volume dose ≥50 Gy), the median dose difference was -1.12 to 0.57 Gy (absolute difference of 1.13-3.25%) for the mean dose to heart and chambers; and -2.25 to 4.45 Gy (absolute difference of 0.94-6.79%) for the mean dose to substructures.

CONCLUSION

The novel hybrid automatic segmentation tool reported high accuracy and consistency over a validation set with challenging anatomical and imaging variations. This has promising applications in substructure dose calculations of large-scale datasets and for future studies on long-term cardiac toxicity.

Machine learning analysis of chest CT to detect cardiac abnormalities in COVID19

Introduction

Standard, un-gated chest CT can be used as the basis of detailed segmentation of the atrial and ventricular cardiac chambers. In conditions such as COVID19 where dedicated cardiac imaging may be hazardous or unavailable atlas-based machine learning tools allow automatic quantification of cardiac morphology and may allow early detection of abnormalities.

Purpose

To develop an automated screening tool to detect cardiac changes associated with COVID19 on chest/lung CT to allow early treatment and appropriate selection of patients for dedicated cardiac imaging.

Methods

A previously validated atlas-based cardiac contouring algorithm was modified to work within the setting of variable and severe lung pathology. The modified technique was used to segment the left and right atria and ventricles from non-contrast CT scans. We applied the developed algorithm to the Moscow University COVID19 CT dataset. 1110 scans were available. COVID19 severity was graded 0 to 4. Grade 4 was not used in analysis due to insufficient numbers. Cardiac chamber sizes were compared according to COVID19 severity status. In a limited cohort of repeat studies, the feasibility of polar mapping to demonstrated serial morphological change was tested.

Results

A statistically significant increase of average cardiac chamber volumes was noted relative to mild Grade 0 COVID19 at every incremental severity grade (Figure 1). Changes in average ventricular volumes were greater (up to 15.2% and 16.9% for left and right ventricles) than changes in atrial volumes (up 12.1% and 7.6% for left and right atria). Automated quantification was successful in the large majority of cases and inter-patient polar mapping of sequential data to detect progressive chamber enlargement appears feasible (Figure 2).

Conclusion

Machine learning methods permit automatic quantification of cardiac chamber size from standard lung CT scans. Cardiac changes on lung CT examinations may be used to identify cardiac abnormalities at an early stage and could be useful to triage individuals for dedicated cardiac investigations. With further refinement, this method may be useful to detect and track temporal cardiac changes in COVID19, as well as in other pulmonary pathology and conditions in which chest CT is routinely used.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): SPHERE Research consortium Figure 1Figure 2

Regional cardiac dysfunction determined by radiation dose in patients with breast cancer

Background

Subclinical left ventricular (LV) dysfunction by 2D global longitudinal strain (GLS) immediately following radiotherapy (RT) and persisting at 12 months has been described in breast cancer (BC) patients. We hypothesised that persistent LV dysfunction may be regional and correlate with segmental RT.

Methods

Transthoracic echocardiograms were performed at baseline, 6 weeks and 12 months post-RT on 61 chemotherapy-naïve women with left sided BC.

Results

Anterior and anteroseptal regions received the highest radiation dose, and posterior and inferior segments the lowest radiation dose (Figure 1). Within each region, there was a progressive increase in the radiation dose received from base to apex. At 6 weeks, the greatest reduction in strain was observed in the anterior and anteroseptal regions, with the most significant reduction in strain in the apical segments. At 12 months, despite improvement in strain, the percentage reduction in strain was similar. There was a significant interaction between both region and segment, on the percentage change in strain at 6 weeks (p<0.001) and at 12 months (p=0.007). Pairwise comparisons of apical to basal percentage change in strain demonstrated the most significant reductions in the anterior region at 6 weeks and 12 months (Table 1).

Conclusions

RT causes segmental myocardial dysfunction, with areas receiving the highest RT demonstrating the largest impairment in strain, with these changes persisting at 12 months. Long term correlation with adverse events is required.

Figure 1. Radiation dose by region and segment

Funding Acknowledgement

Type of funding source: None

Cardiac magnetic resonance derived left atrial function after ST-elevation myocardial infarction: an important prognostic indicator

Background

The prognostic value of cardiac magnetic resonance (CMR) derived left atrial (LA) strain, ejection fraction (LAEF) and volumes (LAVImax and LAVImin) after STEMI is controversial.

Aim

To assess the relationship between LA function and major adverse cardiovascular events (MACE) within 2 years after STEMI.

Methods

We prospectively recruited 213 consecutive STEMI patients who underwent CMR at median day 4. 202 patients had complete CMR data for feature tracking assessment. LA reservoir and booster strain were quantified by one blinded observer based on the average of three independently repeated measurements from two- and four-chamber views. MACE was a composite of all-cause mortality, reinfarction, new or worsening heart failure, stroke and sustained ventricular arrhythmias.

Results

The cohort included 174 (86.1%) males, median age 56 years (IQR 50–65 years). MACE occurred in 35 (17.3%) patients. Patients with MACE had lower median reservoir strain (18.9 vs 29.4%, p<0.001), booster strain (9.4 vs 13.0%, p=0.002) and LAEF (41.5 vs 49.2%, p<0.001), and higher LAVImax (43.5 vs 38.6ml/m2, p=0.019) and LAVImin (23.7 vs 19.3ml/m2, p<0.001) than patients without MACE. Patients with reduced left ventricular ejection fraction (LVEF≤40%) had lower median reservoir strain (22.5 vs 30.1%, p<0.001), booster strain (11.3 vs 12.9%, p=0.021) and LAEF (43.3 vs 50.3%, p<0.001) than patients with LVEF>40%. AUC analyses showed reservoir strain (AUC 0.769; 95% CI 0.676–0.861, p<0.001), booster strain (AUC 0.684; 95% CI 0.558–0.810, p=0.002) and LAEF (AUC 0.698; 95% CI 0.596–0.800, p<0.001) predicted MACE. Kaplan Meier analyses showed a difference in MACE between high- and low-risk groups for reservoir strain (cutoff 21%, p<0.001), booster strain (cutoff 9.6%, p<0.001) and LAEF (cutoff 41%, p<0.001). Univariate Cox regression analyses showed all LA parameters had a significant effect on MACE, while multivariate analyses found additional prognostic utility using reservoir strain.

Conclusion

LA reservoir strain provided incremental prognostic value beyond established clinical and CMR parameters for predicting MACE after STEMI.

Kaplan Meier analyses

Funding Acknowledgement

Type of funding source: None

4938Left ventricular global longitudinal strain recovery predicts scar size reduction and systolic remodelling post ST-elevation myocardial infarction

Background

Left ventricular (LV) strain has prognostic utility following ST-elevation myocardial infarction (STEMI); however, serial changes in LV strain has not been evaluated post-infarct. We sought to determine the relationship between post-STEMI transthoracic echocardiographic (TTE) LV global longitudinal strain (GLS) and cardiac magnetic resonance (CMR) imaging derived scar size and LV systolic remodelling.

Methods

Following revascularisation, 172 first STEMI patients (85% male, 56.9±10.7 years) had paired TTE for GLS, and CMR to evaluate scar size and LV systolic function at baseline (2–7 days) and follow-up (8–10 weeks). Patients were divided into 3 groups according to absolute baseline GLS: group 1 (GLS ≥16%), group 2 (12%< GLS <16%), group 3 (GLS ≤12%). GLS recovery was defined as ≥10% increase in GLS at follow-up, excluding patients with normal baseline GLS. LV systolic adverse remodelling was defined as ≥15% increase in LVESV. LV systolic reverse remodelling was defined as ≥15% decrease in LVESV. Scar reduction was defined as ≥30% decrease in scar size.

Results

Group 1 and 2 had smaller follow-up scar size and higher LVEF compared to group 3 (p<0.0001 for both, see table). There was no difference in scar size reduction or systolic reverse remodeling among the baseline GLS groups (p>0.05 for both). Importantly, no patients from group 1 demonstrated systolic adverse remodelling. Relative change in GLS is significantly correlated with changes in LVEF (r=0.354, p<0.0001) and scar size (r=−0.262, p<0.0001), see figure. On multivariate binary logistic analysis, patients who demonstrated GLS recovery had greater reduction in scar size (OR=2.77 (1.09–7.01), p=0.032) and LV systolic reverse remodelling (OR=9.63 (1.21–76.41), p=0.032).

Follow-up parameters within GLS groups All patients (n=172) Group 1 (n=47) Group 2 (n=72) Group 3 (n=53) Follow-up GLS, % 16.02±3.44 19.38±1.90 16.36±2.09 12.57±2.69 GLS recovery, n 110 (64%) 19 (40%) 53 (74%) 38 (72%) Follow-up scar size, % 7.67±5.40 5.01±3.38 6.27±3.73 12.02±6.24 Follow-up LVEF, % 51.80±10.20 57.83±6.95 54.14±8.02 43.26±9.83 Data presented as mean ± SD or n (%).

Correlation graphs for change in GLS

Conclusion

Stratification of STEMI patients by baseline GLS was a determinant of CMR scar size as well as LV systolic function. However, the evaluation of GLS recovery could provide additional insights into reduction in scar size and LV systolic remodelling, both important prognostic markers. Thus, echocardiographic serial GLS evaluation may be a relevant non-invasive parameter, that is cheaper and more widely available for monitoring STEMI patients and guiding therapy.

P1483Serial changes in peak left atrial strain predicts diastolic remodelling following percutaneous revascularisation for ST-segment elevation myocardial infarction

Background

Echocardiographic 2D speckle tracking peak left atrial (LA) strain reflects LA reservoir function. Limited studies have reported the relationship between peak LA strain and diastolic dysfunction. In addition, adverse diastolic remodeling (ADR) has been reported, to have better prognostic value than single diastolic function assessment following ST-elevation myocardial infarction (STEMI).

Purpose

We examined the relationship between serial changes in echocardiographic peak LA strain and diastolic function in STEMI patients.

Methods

186 percutaneously revascularized first presentation STEMI patients (87% male, 56.9±10.6 years) underwent serial TTE at baseline (2–7 days) and at follow-up (8–10 weeks) post-STEMI. Peak LA reservoir strain measurements were analysed from apical 2-, 3- and 4- chamber views. Diastolic function was graded as per 2016 guidelines: normal, grade 1, grade 2 and grade 3. ADR was defined as worsening of diastolic function grade (≥1) from baseline to follow-up, or persistent grade 3.

Results

Lower baseline peak LA strain was associated with grade 2 and grade 3 diastolic dysfunction compared to normal and grade 1 function at follow-up (p<0.05, see figure and table). Change in LA strain was less with worsening grades of diastolic function (see table). ADR was seen in 33 patients. Lower baseline peak LA strain predicted ADR (B=0.86 (0.80–0.92), p<0.0001). In addition, a reduction in peak LA strain at follow up was independently associated with ADR (B=0.91 (0.84–0.97), p=0.007) (see figure).

Diastolic function grades with LA strain Follow-up diastolic function Baseline peak LA strain (%) Follow-up peak LA strain (%) Change in peak LA strain (%) Normal (n=91) 35.70±6.38 40.97±8.10 5.26±6.07 Grade 1 (n=61) 30.70±6.97 34.54±8.77 3.84±6.32 Grade 2 (n=17) 23.95±6.11 25.49±5.93 1.54±4.70 Grade 3 (n=17) 23.66±8.07 23.49±10.64 −0.17±7.44 Data presented as mean ± SD.

Peak LA strain vs diastolic function

Conclusion

Peak LA strain is associated with diastolic function following STEMI and differentiates normal diastolic function from diastolic dysfunction. Serial changes in peak LA strain correlated with diastolic remodelling. Longer-term follow-up is required to determine the prognostic value of changes in peak LA strain, and diastolic remodelling.