A probabilistic deep motion model for unsupervised cardiac shape anomaly assessment

Automatic shape anomaly detection in large-scale imaging data can be useful for screening suboptimal segmentations and pathologies altering the cardiac morphology without intensive manual labour. We propose a deep probabilistic model for local anomaly detection in sequences of heart shapes, modelled as point sets, in a cardiac cycle. A deep recurrent encoder-decoder network captures the spatio-temporal dependencies to predict the next shape in the cycle and thus derive the outlier points that are attributed to excessive deviations from the network prediction. A predictive mixture distribution models the inlier and outlier classes via Gaussian and uniform distributions, respectively. A Gibbs sampling Expectation-Maximisation (EM) algorithm computes soft anomaly scores of the points via the posterior probabilities of each class in the E-step and estimates the parameters of the network and the predictive distribution in the M-step. We demonstrate the versatility of the method using two shape datasets derived from: (i) one million biventricular CMR images from 20,000 participants in the UK Biobank (UKB), and (ii) routine diagnostic imaging from Multi-Centre, Multi-Vendor, and Multi-Disease Cardiac Image (M&Ms). Experiments show that the detected shape anomalies in the UKB dataset are mostly associated with poor segmentation quality, and the predicted shape sequences show significant improvement over the input sequences. Furthermore, evaluations on U-Net based shapes from the M&Ms dataset reveals that the anomalies are attributable to the underlying pathologies that affect the ventricles. The proposed model can therefore be used as an effective mechanism to sift shape anomalies in large-scale cardiac imaging pipelines for further analysis.

Unsupervised Anomaly Detection in Multivariate Spatio-Temporal Data Using Deep Learning: Early Detection of COVID-19 Outbreak in Italy

Unsupervised anomaly detection for spatio-temporal data has extensive use in a wide variety of applications such as earth science, traffic monitoring, fraud and disease outbreak detection. Most real-world time series data have a spatial dimension as an additional context which is often expressed in terms of coordinates of the region of interest (such as latitude - longitude information). However, existing techniques are limited to handle spatial and temporal contextual attributes in an integrated and meaningful way considering both spatial and temporal dependency between observations. In this paper, a hybrid deep learning framework is proposed to solve the unsupervised anomaly detection problem in multivariate spatio-temporal data. The proposed framework works with unlabeled data and no prior knowledge about anomalies are assumed. As a case study, we use the public COVID-19 data provided by the Italian Department of Civil Protection. Northern Italy regions' COVID-19 data are used to train the framework; and then any abnormal trends or upswings in COVID-19 data of central and southern Italian regions are detected. The proposed framework detects early signals of the COVID-19 outbreak in test regions based on the reconstruction error. For performance comparison, we perform a detailed evaluation of 15 algorithms on the COVID-19 Italy dataset including the state-of-the-art deep learning architectures. Experimental results show that our framework shows significant improvement on unsupervised anomaly detection performance even in data scarce and high contamination ratio scenarios (where the ratio of anomalies in the data set is more than 5%). It achieves the earliest detection of COVID-19 outbreak and shows better performance on tracking the peaks of the COVID-19 pandemic in test regions. As the timeliness of detection is quite important in the fight against any outbreak, our framework provides useful insight to suppress the resurgence of local novel coronavirus outbreaks as early as possible.

Heavy metal-induced stress in rice crops detected using multi-temporal Sentinel-2 satellite images

Regional-level information on heavy metal pollution in agro-ecosystems is essential for food security because excessive levels of heavy metals in crops may pose risks to humans. However, collecting this information over large areas is inherently costly. This paper investigates the possibility of applying multi-temporal Sentinel-2 satellite images to detect heavy metal-induced stress (i.e., Cd stress) in rice crops in four study areas in Zhuzhou City, Hunan Province, China. For this purpose, we compared seven Sentinel-2 images acquired in 2016 and 2017 with in situ measured hyper-spectral data, chlorophyll content, rice leaf area index, and heavy metal concentrations in soil collected from 2014 to 2017. Vegetation indices (VIs) related to red edge bands were referred to as the sensitive indicators for screening stressed rice from unstressed rice. The coefficients of spatio-temporal variation (CSTV) derived from the VIs allowed us to discriminate crops exposed to pollution from heavy metals as well as environmental stressors. The results indicate that (i) the red edge chlorophyll index, the red edge position index, and the normalized difference red edge 2 index derived from multi-temporal Sentinel-2 images were good indicators for screening stressed rice from unstressed rice; (ii) Rice under Cd stress remained stable with lower CSTV values of VIs overall growth stages in the experimental region, whereas rice under other stressors (i.e., pests and disease) showed abrupt changes at some growth stages and presented "hot spots" with greater CSTV values; and (iii) the proposed spatio-temporal anomaly detection method was successful at detecting rice under Cd stress; and CSTVs of rice VIs stabilized regardless of whether they were applied to consecutive growth stages or to two different crop years. This study suggests that regional heavy metal stress may be accurately detected using multi-temporal Sentinel-2 images, using VIs sensitive to the spatio-temporal characteristics of crops.