FULLY AUTOMATIC SEGMENTATION OF THE RIGHT VENTRICLE VIA MULTI-TASK DEEP NEURAL NETWORKS

Deep-active-learning approach towards accurate right ventricular segmentation using a two-level uncertainty estimation

The Right Ventricle (RV) is currently recognised to be a significant and important prognostic factor for various pathologies. Its assessment is made possible using Magnetic Resonance Imaging (CMRI) short-axis slices. Yet, due to the challenging issues of this cavity, radiologists still perform its delineation manually, which is tedious, laborious, and time-consuming. Therefore, to automatically tackle the RV segmentation issues, Deep-Learning (DL) techniques seem to be the axis of the most recent promising approaches. Along with its potential at dealing with shape variations, DL techniques highly requires a large number of labelled images to avoid over-fitting. Subsequently, with the produced large amounts of data in the medical industry, preparing annotated datasets manually is still time-consuming, and requires high skills to be accomplished. To benefit from a significant number of labelled and unlabelled CMRI images, a Deep-Active-Learning (DAL) approach is proposed in this paper to segment the RV. Thus, three main steps are distinguished. First, a personalised labelled dataset is gathered and augmented to allow initial learning. Secondly, a U-Net based architecture is modified towards efficient initial accuracy. Finally, a two-level uncertainty estimation technique is settled to enable the selection of complementary unlabelled data. The proposed pipeline is enhanced with a customised postprocessing step, in which epistemic uncertainty and Dense Conditional Random Fields are used. The proposed approach is tested on 791 images gathered from 32 public patients and 1230 images of 50 custom subjects. The obtained results show an increased dice coefficient from 0.86 to 0.91 with a decreased Hausdorff distance from 7.55 to 7.45.

Deep learning can yield clinically useful right ventricular segmentations faster than fully manual analysis

Right ventricular (RV) volumes are commonly obtained through time-consuming manual delineations of cardiac magnetic resonance (CMR) images. Deep learning-based methods can generate RV delineations, but few studies have assessed their ability to accelerate clinical practice. Therefore, we aimed to develop a clinical pipeline for deep learning-based RV delineations and validate its ability to reduce the manual delineation time. Quality-controlled delineations in short-axis CMR scans from 1114 subjects were used for development. Time reduction was assessed by two observers using 50 additional clinical scans. Automated delineations were subjectively rated as (A) sufficient for clinical use, or as needing (B) minor or (C) major corrections. Times were measured for manual corrections of delineations rated as B or C, and for fully manual delineations on all 50 scans. Fifty-eight % of automated delineations were rated as A, 42% as B, and none as C. The average time was 6 min for a fully manual delineation, 2 s for an automated delineation, and 2 min for a minor correction, yielding a time reduction of 87%. The deep learning-based pipeline could substantially reduce the time needed to manually obtain clinically applicable delineations, indicating ability to yield right ventricular assessments faster than fully manual analysis in clinical practice. However, these results may not generalize to clinics using other RV delineation guidelines.

Cardiac phase-resolved late gadolinium enhancement imaging

Late gadolinium enhancement (LGE) with cardiac magnetic resonance (CMR) imaging is the clinical reference for assessment of myocardial scar and focal fibrosis. However, current LGE techniques are confined to imaging of a single cardiac phase, which hampers assessment of scar motility and does not allow cross-comparison between multiple phases. In this work, we investigate a three step approach to obtain cardiac phase-resolved LGE images: (1) Acquisition of cardiac phase-resolved imaging data with varying T₁ weighting. (2) Generation of semi-quantitative T1* maps for each cardiac phase. (3) Synthetization of LGE contrast to obtain functional LGE images. The proposed method is evaluated in phantom imaging, six healthy subjects at 3T and 20 patients at 1.5T. Phantom imaging at 3T demonstrates consistent contrast throughout the cardiac cycle with a coefficient of variation of 2.55 ± 0.42%. In-vivo results show reliable LGE contrast with thorough suppression of the myocardial tissue is healthy subjects. The contrast between blood and myocardium showed moderate variation throughout the cardiac cycle in healthy subjects (coefficient of variation 18.2 ± 3.51%). Images were acquired at 40–60 ms and 80 ms temporal resolution, at 3T and 1.5, respectively. Functional LGE images acquired in patients with myocardial scar visualized scar tissue throughout the cardiac cycle, albeit at noticeably lower imaging resolution and noise resilience than the reference technique. The proposed technique bears the promise of integrating the advantages of phase-resolved CMR with LGE imaging, but further improvements in the acquisition quality are warranted for clinical use.

Hybrid Deep Feature Fusion of 2D CNN and 3D CNN for Vestibule Segmentation from CT Images

The accurate vestibule segmentation from CT images is essential to the quantitative analysis of the anatomical structure of the ear. However, it is a challenging task due to the tiny size, blur boundary, and drastic variations in shape and size. In this paper, according to the specific characteristics and segmentation requirements of the vestibule, a vestibule segmentation network with a hybrid deep feature fusion of 2D CNN and 3D CNN is proposed. First, a 2D CNN is designed to extract the intraslice features through multiple deep feature fusion strategies, including a convolutional feature fusion strategy for different receptive fields, a feature channel fusion strategy based on channel attention mechanism, and an encoder-decoder feature fusion strategy. Next, a 3D DenseUNet is designed to extract the interslice features. Finally, a hybrid feature fusion module is proposed to fuse the intraslice and interslice features to effectively exploit the context information, thus achieving the accurate segmentation of the vestibule structure. At present, there is no publicly available dataset for vestibule segmentation. Therefore, the proposed segmentation method is validated on two self-established datasets, namely, VestibuleDataSet and IEBL-DataSet. It has been compared with several state-of-the-art methods on the datasets, including the general DeeplabV3+ method and specific 3D DSD vestibule segmentation method. The experimental results show that our proposed method can achieve superior segmentation accuracy.

Assessment of right ventricular size and function from cardiovascular magnetic resonance images using artificial intelligence

BACKGROUND

Theoretically, artificial intelligence can provide an accurate automatic solution to measure right ventricular (RV) ejection fraction (RVEF) from cardiovascular magnetic resonance (CMR) images, despite the complex RV geometry. However, in our recent study, commercially available deep learning (DL) algorithms for RVEF quantification performed poorly in some patients. The current study was designed to test the hypothesis that quantification of RV function could be improved in these patients by using more diverse CMR datasets in addition to domain-specific quantitative performance evaluation metrics during the cross-validation phase of DL algorithm development.

METHODS

We identified 100 patients from our prior study who had the largest differences between manually measured and automated RVEF values. Automated RVEF measurements were performed using the original version of the algorithm (DL1), an updated version (DL2) developed from a dataset that included a wider range of RV pathology and validated using multiple domain-specific quantitative performance evaluation metrics, and conventional methodology performed by a core laboratory (CORE). Each of the DL-RVEF approaches was compared against CORE-RVEF reference values using linear regression and Bland-Altman analyses. Additionally, RVEF values were classified into 3 categories: ≤ 35%, 35-50%, and ≥ 50%. Agreement between RVEF classifications made by the DL approaches and the CORE measurements was tested.

RESULTS

CORE-RVEF and DL-RVEFs were obtained in all patients (feasibility of 100%). DL2-RVEF correlated with CORE-RVEF better than DL1-RVEF (r = 0.87 vs. r = 0.42), with narrower limits of agreement. As a result, DL2 algorithm also showed increasing accuracy from 0.53 to 0.80 for categorizing RV function.

CONCLUSIONS

The use of a new DL algorithm cross-validated on a dataset with a wide range of RV pathology using multiple domain-specific metrics resulted in a considerable improvement in the accuracy of automated RVEF measurements. This improvement was demonstrated in patients whose images were the most challenging and resulted in the largest RVEF errors. These findings underscore the critical importance of this strategy in the development of DL approaches for automated CMR measurements.

Deep Learning for Cardiac Image Segmentation: A Review

Deep learning has become the most widely used approach for cardiac image segmentation in recent years. In this paper, we provide a review of over 100 cardiac image segmentation papers using deep learning, which covers common imaging modalities including magnetic resonance imaging (MRI), computed tomography (CT), and ultrasound and major anatomical structures of interest (ventricles, atria, and vessels). In addition, a summary of publicly available cardiac image datasets and code repositories are included to provide a base for encouraging reproducible research. Finally, we discuss the challenges and limitations with current deep learning-based approaches (scarcity of labels, model generalizability across different domains, interpretability) and suggest potential directions for future research.