Uncertainty Estimation in Medical Image Denoising with Bayesian Deep Image Prior

Laves, Max-Heinrich; Tölle, Malte; Ortmaier, Tobias

doi:10.1007/978-3-030-60365-6_9

Cited by 30 publications

(43 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, due to the lack of ground truth, we were unable to make concrete conclusions about the performance of our CNN on test patient data. But the uncertainty of our CNN can be quantified by generating confidence maps 34–36 using Bayesian networks, 37 an ensemble of multiple networks, 38 or an extension of the probabilistic U‐Net, 39 which can be one direction to investigate in the future.…”

Section: Discussionmentioning

confidence: 99%

DblurDoseNet: A deep residual learning network for voxel radionuclide dosimetry compensating for single‐photon emission computerized tomography imaging resolution

Fessler

Mikell

et al. 2021

Medical Physics

View full text Add to dashboard Cite

Purpose: Current methods for patient specific voxel-level dosimetry in radionuclide therapy suffer from a trade-off between accuracy and computational efficiency. Monte Carlo (MC) radiation transport is considered as the gold standard but is computationally expensive, while faster dose voxel kernel (DVK) convolution can be sub-optimal in the presence of tissue heterogeneities.Furthermore, the accuracies of both these methods are limited by the spatial resolution of the reconstructed emission image. To overcome these limitations, this paper takes a novel approach of constructing a single deep convolutional neural network (CNN) named as DblurDoseNet that learns to produce dose-rate maps while compensating for the limited resolution of SPECT images. Methods:To mitigate the effects of poor SPECT resolution and reconstruction artifacts on dosimetry, we trained our CNN using MC-generated dose-rate maps that directly corresponded to the true activity maps in virtual patient phantoms. We applied residual learning such that our CNN only learned the difference between the true dose-rate map and DVK dose-rate map with density scaling. The network consists of a depth feature extractor and a 2D U-Net, where the input was 11 slices (3.3 cm) of Lu-177 SPECT/CT images and the output was the dose-rate map corresponding to the center slice. In addition to phantoms, 42 SPECT/CT scans of patients who underwent Lu-177 DOTATATE therapy were also used for testing. Results: In test phantoms, the lesion/organ mean dose-rate error and the normalized root mean square error (NRMSE) relative to ground-truth for the CNN method was consistently lower than DVK and MC. In particular, for CNN compared to DVK/MC, the average improvement in mean dose error was 55%/53% and 66%/56%; and in NRMSE was 18%/17% and 10%/11% for lesion and kidney, respectively. Line profiles and dose-volume histograms demonstrated compensation for SPECT resolution effects in the CNN generated dose-rate maps. Noise, determined from multiple Poisson realizations, showed an average improvement of 21%/27% compared to DVK/MC. In patients, a high concordance was observed between CNN and MC in joint histogram analysis. The trained residual CNN took ~30 seconds on GPU to generate a (512 × 512 × 130) dose-rate map for a patient. Conclusion: The proposed CNN is well-suited for real-time patient-specific dosimetry for clinical treatment planning due to its demonstrated improvement in accuracy, resolution, noise and speed over the current gold-standard.

show abstract

Section: Discussionmentioning

confidence: 99%

DblurDoseNet: A deep residual learning network for voxel radionuclide dosimetry compensating for single‐photon emission computerized tomography imaging resolution

Fessler

Mikell

et al. 2021

Medical Physics

View full text Add to dashboard Cite

show abstract

“…SGLD DIP has already been applied to PET image reconstruction (Carrillo et al, 2021). Prior to (Tölle et al, 2021), we have shown that DIP with SGLD shows almost unchanged overfitting behavior in the case of medical images (Laves et al, 2020b). As a solution, we proposed a variational inference (VI) approach to DIP using Monte Carlo dropout (MCD) (Gal and Ghahramani, 2016).…”

Section: Related Workmentioning

confidence: 99%

“…These artifacts are composed out of learned image statistics, which can lead to false anatomical structures being embedded in the reconstruction that are not present in the imaged object (Bhadra et al, 2020). This phenomenon is referred to as hallucination and is not limited to tomographic reconstruction but also happens in other deeplearning-based inverse image tasks (Laves et al, 2020b). Hallucinations can result in misdiagnosis and must be avoided at all costs in medical imaging.…”

Section: Introductionmentioning

confidence: 99%

Posterior temperature optimized Bayesian models for inverse problems in medical imaging

Laves¹,

Tölle²,

Schlaefer³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

We present Posterior Temperature Optimized Bayesian Inverse Models (POTOBIM), an unsupervised Bayesian approach to inverse problems in medical imaging using mean-field variational inference with a fully tempered posterior. Bayesian methods exhibit useful properties for approaching inverse tasks, such as tomographic reconstruction or image denoising. A suitable prior distribution introduces regularization, which is needed to solve the ill-posed problem and reduces overfitting the data. In practice, however, this often results in a suboptimal posterior temperature, and the full potential of the Bayesian approach is not being exploited. In POTOBIM, we optimize both the parameters of the prior distribution and the posterior temperature with respect to reconstruction accuracy using Bayesian optimization with Gaussian process regression. Our method is extensively evaluated on four different inverse tasks on a variety of modalities with images from public data sets and we demonstrate that an optimized posterior temperature outperforms both non-Bayesian and Bayesian approaches without temperature optimization. The use of an optimized prior distribution and posterior temperature leads to improved accuracy and uncertainty estimation and we show that it is sufficient to find these hyperparameters per task domain. Well-tempered posteriors yield calibrated uncertainty, which increases the reliability in the predictions. Our source code is publicly available at github.com/Cardio-AI/mfvi-dip-mia.

show abstract

“…This is a commonly observed scenario in medical imaging applications due to variations among patients, image acquisition, and reconstruction protocols [8,9,12]. For example, when applying denoising ConvNets on unseen features, it may cause artifacts in the denoised images as demonstrated in both Ultrasound [14] and Positron Emission Tomography (PET) applications [5]. To generalize a trained ConvNet to different image distributions, one has to include images sampled from the new distribution (task) in the training dataset and retrain the ConvNet.…”

Section: Introductionmentioning

confidence: 99%

Targeted Gradient Descent: A Novel Method for Convolutional Neural Networks Fine-tuning and Online-learning

Chen,

Asma,

Chan

2021

Preprint

View full text Add to dashboard Cite

A convolutional neural network (ConvNet) is usually trained and then tested using images drawn from the same distribution. To generalize a ConvNet to various tasks often requires a complete training dataset that consists of images drawn from different tasks. In most scenarios, it is nearly impossible to collect every possible representative dataset as a priori. The new data may only become available after the ConvNet is deployed in clinical practice. ConvNet, however, may generate artifacts on out-of-distribution testing samples. In this study, we present Targeted Gradient Descent (TGD), a novel fine-tuning method that can extend a pre-trained network to a new task without revisiting data from the previous task while preserving the knowledge acquired from previous training. To a further extent, the proposed method also enables online learning of patientspecific data. The method is built on the idea of reusing a pre-trained ConvNet's redundant kernels to learn new knowledge. We compare the performance of TGD to several commonly used training approaches on the task of Positron emission tomography (PET) image denoising. Results from clinical images show that TGD generated results on par with training-from-scratch while significantly reducing data preparation and network training time. More importantly, it enables online learning on the testing study to enhance the network's generalization capability in real-world applications.

show abstract

Uncertainty Estimation in Medical Image Denoising with Bayesian Deep Image Prior

Cited by 30 publications

References 18 publications

DblurDoseNet: A deep residual learning network for voxel radionuclide dosimetry compensating for single‐photon emission computerized tomography imaging resolution

DblurDoseNet: A deep residual learning network for voxel radionuclide dosimetry compensating for single‐photon emission computerized tomography imaging resolution

Posterior temperature optimized Bayesian models for inverse problems in medical imaging

Targeted Gradient Descent: A Novel Method for Convolutional Neural Networks Fine-tuning and Online-learning

Contact Info

Product

Resources

About