Semantic denoising autoencoders for retinal optical coherence tomography

Laves, Max-Heinrich; Ihler, Sontje; Kahrs, Lüder A.; Ortmaier, Tobias

doi:10.1117/12.2526936

Cited by 6 publications

(2 citation statements)

References 10 publications

(7 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…σ ‐scaling, introduced by Laves et al. (2021), uses the validation set to “check” the validity of uncertainty estimates. This check provides a scaling factor based on the results and adjusts uncertainty estimates based on over or under prediction.…”

Section: Methodsmentioning

confidence: 99%

Probabilistic Short‐Term Solar Driver Forecasting With Neural Network Ensembles

Daniell,

Mehta

2024

Space Weather

View full text Add to dashboard Cite

Space weather indices are used to drive forecasts of thermosphere density, which directly affects objects in low‐Earth orbit (LEO) through atmospheric drag force. A set of proxies and indices (drivers), F10.7, S10.7, M10.7, and Y10.7 are used as inputs by the JB2008, (https://doi.org/10.2514/6.2008‐6438) thermosphere density model. The United States Air Force (USAF) operational High Accuracy Satellite Drag Model (HASDM), relies on JB2008, (https://doi.org/10.2514/6.2008‐6438), and forecasts of solar drivers from a linear algorithm. We introduce methods using long short‐term memory (LSTM) model ensembles to improve over the current prediction method as well as a previous univariate approach. We investigate the usage of principal component analysis (PCA) to enhance multivariate forecasting. A novel method, referred to as striped sampling, is created to produce statistically consistent machine learning data sets. We also investigate forecasting performance and uncertainty estimation by varying the training loss function and by investigating novel weighting methods. Results show that stacked neural network model ensembles make multivariate driver forecasts which outperform the operational linear method. When using MV‐MLE (multivariate multi‐lookback ensemble), we see an improvement of RMSE for F10.7, S10.7, M10.7, and Y10.7 of 17.7%, 12.3%, 13.8%, 13.7% respectively, over the operational method. We provide the first probabilistic forecasting method for S10.7, M10.7, and Y10.7. Ensemble approaches are leveraged to provide a distribution of predicted values, allowing an investigation into robustness and reliability (R&R) of uncertainty estimates. Uncertainty was also investigated through the use of calibration error score (CES), with the MV‐MLE providing an average CES of 5.63%, across all drivers.

show abstract

Section: Methodsmentioning

confidence: 99%

Probabilistic Short‐Term Solar Driver Forecasting With Neural Network Ensembles

Daniell,

Mehta

2024

Space Weather

View full text Add to dashboard Cite

show abstract

“…Noise in medical imaging affects all modalities, including X-ray, magnetic resonance imaging (MRI), computed tomography (CT), ultrasound (US) or optical coherence tomography (OCT) and can obstruct important details for medical diagnosis [7,1,16]. Besides "classical" approaches with linear and non-linear filters, such as the Wiener filter, or wavelet-denoising [3,22], convolutional neural networks (CNN) have proven to yield superior performance in denoising of natural and medical images [28,16].…”

Section: Introductionmentioning

confidence: 99%

Uncertainty Estimation in Medical Image Denoising with Bayesian Deep Image Prior

Laves¹,

Tölle²,

Ortmaier³

2020

Preprint

Self Cite

View full text Add to dashboard Cite

Uncertainty quantification in inverse medical imaging tasks with deep learning has received little attention. However, deep models trained on large data sets tend to hallucinate and create artifacts in the reconstructed output that are not anatomically present. We use a randomly initialized convolutional network as parameterization of the reconstructed image and perform gradient descent to match the observation, which is known as deep image prior. In this case, the reconstruction does not suffer from hallucinations as no prior training is performed. We extend this to a Bayesian approach with Monte Carlo dropout to quantify both aleatoric and epistemic uncertainty. The presented method is evaluated on the task of denoising different medical imaging modalities. The experimental results show that our approach yields well-calibrated uncertainty. That is, the predictive uncertainty correlates with the predictive error. This allows for reliable uncertainty estimates and can tackle the problem of hallucinations and artifacts in inverse medical imaging tasks.

show abstract