Search

Atrial fibrillation (AF) is the most common cardiac arrhythmia. It is associated with a higher risk of important adverse health outcomes such as stroke and death. AF is linked to distinct electro-anatomic alterations. The main tool for AF diagnosis is the Electrocardiogram (ECG). However, an ECG recorded at a single time point may not detect individuals with paroxysmal AF. In this study, we developed machine learning models for discrimination of prevalent AF using a combination of image-derived radiomics phenotypes and ECG features. Thus, we characterize the phenotypes of prevalent AF in terms of ECG and imaging alterations. Moreover, we explore sex-differential remodelling by building sex-specific models. Our integrative model including radiomics and ECG together resulted in a better performance than ECG alone, particularly in women. ECG had a lower performance in women than men (AUC: 0.77 vs 0.88, p

Uncertainty estimation in deep learning has become a leading research field in medical image analysis due to the need for safe utilisation of AI algorithms in clinical practice. Most approaches for uncertainty estimation require sampling the network weights multiple times during testing or training multiple networks. This leads to higher training and testing costs in terms of time and computational resources. In this paper, we propose Layer Ensembles, a novel uncertainty estimation method that uses a single network and requires only a single pass to estimate predictive uncertainty of a network. Moreover, we introduce an image-level uncertainty metric, which is more beneficial for segmentation tasks compared to the commonly used pixel-wise metrics such as entropy and variance. We evaluate our approach on 2D and 3D, binary and multi-class medical image segmentation tasks. Our method shows competitive results with state-of-the-art Deep Ensembles, requiring only a single network and a single pass.

The performance of deep segmentation models often degrades due to distribution shifts in image intensities between the training and test data sets. This is particularly pronounced in multi-centre studies involving data acquired using multi-vendor scanners, with variations in acquisition protocols. It is challenging to address this degradation because the shift is often not known \textit{a priori} and hence difficult to model. We propose a novel framework to ensure robust segmentation in the presence of such distribution shifts. Our contribution is three-fold. First, inspired by the spirit of curriculum learning, we design a novel style curriculum to train the segmentation models using an easy-to-hard mode. A style transfer model with style fusion is employed to generate the curriculum samples. Gradually focusing on complex and adversarial style samples can significantly boost the robustness of the models. Second, instead of subjectively defining the curriculum complexity, we adopt an automated gradient manipulation method to control the hard and adversarial sample generation process. Third, we propose the Local Gradient Sign strategy to aggregate the gradient locally and stabilise training during gradient manipulation. The proposed framework can generalise to unknown distribution without using any target data. Extensive experiments on the public M\&Ms Challenge dataset demonstrate that our proposed framework can generalise deep models well to unknown distributions and achieve significant improvements in segmentation accuracy., Accepted by MICCAI-2021

Accurate segmentation of pathological tissue, such as scar tissue and edema, from cardiac magnetic resonance images (CMR) is fundamental to the assessment of the severity of myocardial infarction and myocardial viability. There are many accurate solutions for auto- matic segmentation of cardiac structures from CMR. On the contrary, a solution has not as yet been found for the automatic segmentation of my- ocardial pathological regions due to their challenging nature. As part of the Myocardial Pathology Segmentation combining multi-sequence CMR (MyoPS) challenge, we propose a fully automatic pipeline for segment- ing pathological tissue using registered multi-sequence CMR images se- quences (LGE, bSSFP and T2). The proposed approach involves a two- staged process. First, in order to reduce task complexity, a two-stacked BCDU-net is proposed to a) detect a small ROI based on accurate my- ocardium segmentation and b) perform inside-ROI multi-modal patho- logical region segmentation. Second, in order to regularize the proposed stacked architecture and deal with the under-represented data prob- lem, we propose a synthetic data augmentation pipeline that generates anatomically meaningful samples. The outputs of the proposed stacked BCDU-NET with semantic CMR synthesis are post-processed based on anatomical constrains to re ne output segmentation masks. Results from 25 di erent patients demonstrate that the proposed model improves 1- stage equivalent architectures and bene ts from the addition of synthetic anatomically meaningful samples. A nal ensemble of 15 trained models show a challenge Dice test score of 0.665 0.143 and 0.698 0.128 for scar and scar+edema, respectively.