Exploiting Epistemic Uncertainty of Anatomy Segmentation for Anomaly Detection in Retinal OCT

Abstract

Diagnosis and treatment guidance are aided by detecting relevant biomarkers in medical images. Although supervised deep learning can perform accurate segmentation of pathological areas, it is limited by requiring a-priori definitions of these regions, large-scale annotations, and a representative patient cohort in the training set. In contrast, anomaly detection is not limited to specific definitions of pathologies and allows for training on healthy samples without annotation. Anomalous regions can then serve as candidates for biomarker discovery. Knowledge about normal anatomical structure brings implicit information for detecting anomalies. We propose to take advantage of this property using bayesian deep learning, based on the assumption that epistemic uncertainties will correlate with anatomical deviations from a normal training set. A Bayesian U-Net is trained on a well-defined healthy environment using weak labels of healthy anatomy produced by existing methods. At test time, we capture epistemic uncertainty estimates of our model using Monte Carlo dropout. A novel post-processing technique is then applied to exploit these estimates and transfer their layered appearance to smooth blob-shaped segmentations of the anomalies. We experimentally validated this approach in retinal optical coherence tomography (OCT) images, using weak labels of retinal layers. Our method achieved a Dice index of 0.789 in an independent anomaly test set of age-related macular degeneration (AMD) cases. The resulting segmentations allowed very high accuracy for separating healthy and diseased cases with late wet AMD, dry geographic atrophy (GA), diabetic macular edema (DME) and retinal vein occlusion (RVO). Finally, we qualitatively observed that our approach can also detect other deviations in normal scans such as cut edge artifacts.

Leveraging the Bhattacharyya coefficient for uncertainty quantification in deep neural networks

“…Quantifying uncertainty Variance-based uncertainty quantification using dropout networks is applied to both detection tasks [28,36,50] and segmentation tasks [21,24,34,41,48]. While this metric is most common, alternatives to quantify uncertainty exist, such as the predictive entropy or mutual information [11,18].…”

Deep Learning Approaches for Detecting COVID-19 From Chest X-Ray Images: A Survey

“…In the medical imaging domain, deep learning (DL) techniques have been used to improve the performance of image analysis significantly [9] [10]. For example, DL has been successfully applied to microscopy images [11], brain tumor classification [12], MRI images [13], and retinal photographs [14].…”

An improved agglomerative hierarchical clustering anomaly detection method for scientific data

“…Since the significance of anomalous data, many researchers proposed anomaly detection methods from different application domains and categories. Anomaly detection has been applied in Cyber‐Intrusion Detection, 3 Fraud Detection, 4,5,6 Medical Anomaly Detection, 7 Industrial Damage Detection, 8 Image Processing, 9 Textual Anomaly Detection, 10 Sensor Networks 11 , etc. From the aspect of machine learning, various anomaly detection methods can be classified as supervised, semi‐supervised and unsupervised anomaly detection 2 …”

“…In supervised anomaly detection technologies, support vector machines (SVM) 12 and artificial neural networks 13 perform well, especially for multiclassification detection problems. Semi‐supervised anomaly detection technologies are well‐known for nice outcomes only with normal labeled samples, mainly containing One‐class SVMs 14 and autoencoders 7 . However, the inconvenience of obtaining labeled samples acts as a block of common usage of supervised and semi‐supervised methods.…”

“…Semi-supervised anomaly detection technologies are well-known for nice outcomes only with normal labeled samples, mainly containing One-class SVMs 14 and autoencoders. 7 However, the inconvenience of obtaining labeled samples acts as a block of common usage of supervised and semi-supervised methods.…”