Fully automatic cardiac four chamber and great vessel segmentation on CT pulmonary angiography using deep learning

Abstract

Introduction: Computed tomography pulmonary angiography (CTPA) is an essential test in the work-up of suspected pulmonary vascular disease including pulmonary hypertension and pulmonary embolism. Cardiac and great vessel assessments on CTPA are based on visual assessment and manual measurements which are known to have poor reproducibility. The primary aim of this study was to develop an automated whole heart segmentation (four chamber and great vessels) model for CTPA.

Methods: A nine structure semantic segmentation model of the heart and great vessels was developed using 200 patients (80/20/100 training/validation/internal testing) with testing in 20 external patients. Ground truth segmentations were performed by consultant cardiothoracic radiologists. Failure analysis was conducted in 1,333 patients with mixed pulmonary vascular disease. Segmentation was achieved using deep learning via a convolutional neural network. Volumetric imaging biomarkers were correlated with invasive haemodynamics in the test cohort.

Results: Dice similarity coefficients (DSC) for segmented structures were in the range 0.58-0.93 for both the internal and external test cohorts. The left and right ventricle myocardium segmentations had lower DSC of 0.83 and 0.58 respectively while all other structures had DSC >0.89 in the internal test cohort and >0.87 in the external test cohort. Interobserver comparison found that the left and right ventricle myocardium segmentations showed the most variation between observers: mean DSC (range) of 0.795 (0.785-0.801) and 0.520 (0.482-0.542) respectively. Right ventricle myocardial volume had strong correlation with mean pulmonary artery pressure (Spearman's correlation coefficient = 0.7). The volume of segmented cardiac structures by deep learning had higher or equivalent correlation with invasive haemodynamics than by manual segmentations. The model demonstrated good generalisability to different vendors and hospitals with similar performance in the external test cohort. The failure rates in mixed pulmonary vascular disease were low (<3.9%) indicating good generalisability of the model to different diseases.

Conclusion: Fully automated segmentation of the four cardiac chambers and great vessels has been achieved in CTPA with high accuracy and low rates of failure. DL volumetric biomarkers can potentially improve CTPA cardiac assessment and invasive haemodynamic prediction.

Keywords: computed tomography pulmonary angiography (CTPA); deep-learning (DL); pulmonary vascular disease (PVD); semantic segmentation and labelling; whole heart segmentation.

Figure 1

Methodology. (A) Datasets used within this study. (B) Cardiac localisation deep learning model. (C) Cardiac segmentation deep learning model. (D) Inference pipeline. (E) Model testing strategy. Ob 1, 2, 3, observer 1, 2, 3; DL-1, deep learning model 1; DL-2, deep learning model 2; Dice, dice similarity score, ICC, intraclass correlation coefficient; NSD, normalised surface distance (Surface Dice Score).

Figure 2

Box plots comparing Dice similarity coefficients (DSC) for the segmented cardiac structures for observers 1 (AJS), 2 (KK) and 3 (CJ) and DL model 2 in the interobserver comparison cohort (n = 24). Structures are as follows; LV endocardial cavity (LV_vol), LV myocardium (LV_myo), RV endocardial cavity (RV_vol), RV myocardium (RV_myo), left atrium (LA), right atrium (RA), ascending aorta and aortic arc (Ao_ascend), proximal pulmonary arteries (PA), and descending aorta (Ao_descend).

Figure 3

Box plots comparing Dice similarity coefficient (DSC) for the segmented cardiac structures for DL model 1 vs. the manual segmentation observer 1 (AJS) and DL model 2 vs. observer 1 in the test cohort (n = 100) and the external cohort (n = 20). Structures are as follows; LV endocardial cavity (LV_vol), LV myocardium (LV_myo), RV endocardial cavity (RV_vol), RV myocardium (RV_myo), left atrium (LA), right atrium (RA), ascending aorta and aortic arc (Ao_ascend), proximal pulmonary arteries (PA), and descending aorta (Ao_descend).

Figure 4

Example of a successful segmentation by DL-2 for a patient with suspected PH in the internal test cohort.

Figure 5

Example of a successful segmentation by DL-2 for a patient with suspected PH in the external test cohort.

Figure 6

Bland-Altman plots comparing manually segmented structure volumes by observer 1 (AJS) against DL model 2 in the test cohort (n = 100).

Figure 7

Bland-Altman plots comparing manually segmented structure volumes by observer 1 (AJS) against DL model 2 in the external cohort (n = 20).

Figure 8

(A) Segmentation failure in the LV myocardium in the presence of pericardial effusion. (B) Failure of segmentation of right sided cardiac structures with poor right sided contrast opacification. (C) Segmentation failure apically with globally poor contrast opacification. (D) Example showing success in the presence of an intracardiac device. (E) Segmentation failure in the region of an intracardiac device. (F) Failure in right atrial segmentation with severe dilatation.