Joint Segmentation and Uncertainty Visualization of Retinal Layers in Optical Coherence Tomography Images Using Bayesian Deep Learning

Abstract

Optical coherence tomography (OCT) is commonly used to analyze retinal layers for assessment of ocular diseases. In this paper, we propose a method for retinal layer segmentation and quantification of uncertainty based on Bayesian deep learning. Our method not only performs end-to-end segmentation of retinal layers, but also gives the pixel wise uncertainty measure of the segmentation output. The generated uncertainty map can be used to identify erroneously segmented image regions which is useful in downstream analysis. We have validated our method on a dataset of 1487 images obtained from 15 subjects (OCT volumes) and compared it against the state-of-the-art segmentation algorithms that does not take uncertainty into account. The proposed uncertainty based segmentation method results in comparable or improved performance, and most importantly is more robust against noise.

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

“…In particular, Nair et al [34] used Bayesian supervised learning to segment multiple sclerosis lesions in MRI. Sedai et al [39] applied a similar method for layer segmentation in healthy OCT scans. In both works, aleatoric uncertainty was used for training.…”

“…In both works, aleatoric uncertainty was used for training. In [34], epistemic uncertainty was applied to refine the segmentations, while in [39] the epistemic uncertainty was provided as qualitative feedback to users. Monte Carlo sampling with dropout was used in [36] to average multiple outputs from an autoencoder trained in healthy data.…”

End-to-end forecasting of needle trajectory in percutaneous ablation

“…The performance increase seen in this work is substantially greater than that seen in many works exploring MCDO in segmentation, which either found no improvement 9,10 or incremental performance gains. [11][12][13][14] In addition, the generated uncertainty map highlights potentially incorrect or distorted predictions, allowing an operator to review predictions that may be of inadequate quality.…”

Digitale Bildverarbeitung und Tiefe Neuronale Netze in der Augenheilkunde – aktuelle Trends

“…Auch die Darstellung von Informationen wie z. B. die Quantifizierung der Unsicherheiten des Modells oder eine nachvollziehbare Entscheidungsfindung sind aktive Forschungsgebiete [13]. Weitere Arbeiten beschäftigen sich mit dem Identifizieren bislang unbekannter Merkmale, die ein Indiz für Krankheiten sind [14].…”