scholarly journals Noise2Noise Improved by Trainable Wavelet Coefficients for PET Denoising

Electronics ◽  
2021 ◽  
Vol 10 (13) ◽  
pp. 1529
Author(s):  
Seung-Kwan Kang ◽  
Si-Young Yie ◽  
Jae-Sung Lee

The significant statistical noise and limited spatial resolution of positron emission tomography (PET) data in sinogram space results in the degradation of the quality and accuracy of reconstructed images. Although high-dose radiotracers and long acquisition times improve the PET image quality, the patients’ radiation exposure increases and the patient is more likely to move during the PET scan. Recently, various data-driven techniques based on supervised deep neural network learning have made remarkable progress in reducing noise in images. However, these conventional techniques require clean target images that are of limited availability for PET denoising. Therefore, in this study, we utilized the Noise2Noise framework, which requires only noisy image pairs for network training, to reduce the noise in the PET images. A trainable wavelet transform was proposed to improve the performance of the network. The proposed network was fed wavelet-decomposed images consisting of low- and high-pass components. The inverse wavelet transforms of the network output produced denoised images. The proposed Noise2Noise filter with wavelet transforms outperforms the original Noise2Noise method in the suppression of artefacts and preservation of abnormal uptakes. The quantitative analysis of the simulated PET uptake confirms the improved performance of the proposed method compared with the original Noise2Noise technique. In the clinical data, 10 s images filtered with Noise2Noise are virtually equivalent to 300 s images filtered with a 6 mm Gaussian filter. The incorporation of wavelet transforms in Noise2Noise network training results in the improvement of the image contrast. In conclusion, the performance of Noise2Noise filtering for PET images was improved by incorporating the trainable wavelet transform in the self-supervised deep learning framework.

Entropy ◽  
2021 ◽  
Vol 23 (6) ◽  
pp. 711
Author(s):  
Mina Basirat ◽  
Bernhard C. Geiger ◽  
Peter M. Roth

Information plane analysis, describing the mutual information between the input and a hidden layer and between a hidden layer and the target over time, has recently been proposed to analyze the training of neural networks. Since the activations of a hidden layer are typically continuous-valued, this mutual information cannot be computed analytically and must thus be estimated, resulting in apparently inconsistent or even contradicting results in the literature. The goal of this paper is to demonstrate how information plane analysis can still be a valuable tool for analyzing neural network training. To this end, we complement the prevailing binning estimator for mutual information with a geometric interpretation. With this geometric interpretation in mind, we evaluate the impact of regularization and interpret phenomena such as underfitting and overfitting. In addition, we investigate neural network learning in the presence of noisy data and noisy labels.


Blood ◽  
2003 ◽  
Vol 102 (1) ◽  
pp. 53-59 ◽  
Author(s):  
Karoline Spaepen ◽  
Sigrid Stroobants ◽  
Patrick Dupont ◽  
Peter Vandenberghe ◽  
Johan Maertens ◽  
...  

Abstract The study assessed the prognostic value of fluorine 18-fluorodeoxyglucose positron emission tomography ([18F]FDG-PET) after salvage chemotherapy before high-dose chemotherapy with stem cell transplantation (HDT/SCT) in patients with induction failure or relapsing chemosensitive lymphoma. Retrospective analysis of the clinical and conventional imaging data of 60 patients scheduled for HDT/SCT was performed in parallel with the analysis of the [18F]FDG-PET results. To determine the ability of [18F]FDG-PET to predict clinical outcome, PET images were reread without knowledge of conventional imaging and clinical history. Presence or absence of abnormal [18F]FDG uptake was related to progression-free survival (PFS) and overall survival (OS) using Kaplan-Meier survival analysis. Thirty patients showed a negative [18F]FDG-PET scan before HDT/SCT; 25 of those remained in complete remission, with a median follow-up of 1510 days. Two patients died due to a treatment-related mortality but without evidence of recurrent disease at that time (228-462 days). Only 3 patients had a relapse (median PFS, 1083 days) after a negative [18F]FDG-PET scan. Persistent abnormal [18F]FDG uptake was seen in 30 patients and 26 progressed (median PFS, 402 days); of these 26, 16 died from progressive disease (median OS, 408 days). Four patients are still in complete remission after a positive scan. Comparison between groups indicated a statistically significant association between [18F]FDG-PET findings and PFS (P < .000001) and OS (P < .00002). [18F]FDG-PET has an important prognostic role in the pretransplantation evaluation of patients with lymphoma and enlarges the concept of chemosensitivity used to select patients for HDT/SCT. (Blood. 2003;102:53-59)


2011 ◽  
Vol 65 ◽  
pp. 497-502
Author(s):  
Yan Wei Wang ◽  
Hui Li Yu

A feature matching algorithm based on wavelet transform and SIFT is proposed in this paper, Firstly, Biorthogonal wavelet transforms algorithm is used for medical image to delaminating, and restoration the processed image. Then the SIFT (Scale Invariant Feature Transform) applied in this paper to abstracting key point. Experimental results show that our algorithm compares favorably in high-compressive ratio, the rapid matching speed and low storage of the image, especially for the tilt and rotation conditions.


1999 ◽  
Vol 86 (3) ◽  
pp. 1081-1091 ◽  
Author(s):  
Vincent Pichot ◽  
Jean-Michel Gaspoz ◽  
Serge Molliex ◽  
Anestis Antoniadis ◽  
Thierry Busso ◽  
...  

Heart rate variability is a recognized parameter for assessing autonomous nervous system activity. Fourier transform, the most commonly used method to analyze variability, does not offer an easy assessment of its dynamics because of limitations inherent in its stationary hypothesis. Conversely, wavelet transform allows analysis of nonstationary signals. We compared the respective yields of Fourier and wavelet transforms in analyzing heart rate variability during dynamic changes in autonomous nervous system balance induced by atropine and propranolol. Fourier and wavelet transforms were applied to sequences of heart rate intervals in six subjects receiving increasing doses of atropine and propranolol. At the lowest doses of atropine administered, heart rate variability increased, followed by a progressive decrease with higher doses. With the first dose of propranolol, there was a significant increase in heart rate variability, which progressively disappeared after the last dose. Wavelet transform gave significantly better quantitative analysis of heart rate variability than did Fourier transform during autonomous nervous system adaptations induced by both agents and provided novel temporally localized information.


2020 ◽  
Vol 24 (4) ◽  
pp. 133-145
Author(s):  
E. V. Kryukov ◽  
V. N. Troyan ◽  
O. A. Rukavitsyn ◽  
S. A. Alekseev ◽  
S. I. Kurbanov ◽  
...  

The article presents the possibilities of the complex application of methods of radiation diagnostics: bone x-ray, dual-energy X-ray absorptiometry, computed tomography, positron emission tomography combined with computed tomography using fluorodeoxyglucose labeled with 18-fluorine (PET/CT with 18F-FDG) in a patient with multiple myeloma, which was treated in the amount of high-dose therapy with autologous transplantation of hematopoietic stem cells. The diagnosis was established immunohistochemically. The use of these methods allowed us to dynamically assess the pathological changes characteristic of multiple myeloma.


2011 ◽  
Vol 2011 ◽  
pp. 1-10 ◽  
Author(s):  
Timur Düzenli ◽  
Nalan Özkurt

The performance of wavelet transform-based features for the speech/music discrimination task has been investigated. In order to extract wavelet domain features, discrete and complex orthogonal wavelet transforms have been used. The performance of the proposed feature set has been compared with a feature set constructed from the most common time, frequency and cepstral domain features such as number of zero crossings, spectral centroid, spectral flux, and Mel cepstral coefficients. The artificial neural networks have been used as classification tool. The principal component analysis has been applied to eliminate the correlated features before the classification stage. For discrete wavelet transform, considering the number of vanishing moments and orthogonality, the best performance is obtained with Daubechies8 wavelet among the other members of the Daubechies family. The dual tree wavelet transform has also demonstrated a successful performance both in terms of accuracy and time consumption. Finally, a real-time discrimination system has been implemented using the Daubhecies8 wavelet which has the best accuracy.


Sign in / Sign up

Export Citation Format

Share Document