Unsupervised Medical Image Translation With Adversarial Diffusion Models

Özbey, Muzaffer; Dalmaz, Onat; Dar, Salman UH; Bedel, Hasan A.; Özturk, Şaban; Güngör, Alper; Çukur, Tolga

doi:10.1109/tmi.2023.3290149

Cited by 129 publications

(15 citation statements)

References 64 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…More advanced generative models have recently been researched. Among them, the diffusion model provides state-of-the-art performance, and numerous applications have been implemented in medical imaging (Özbey et al 2023(Özbey et al , Güngör et al 2023. Other generative models should be used to improve the quality of synthetic breast images.…”

Section: Discussionmentioning

confidence: 99%

Power-law spectrum-based objective function to train a generative adversarial network with transfer learning for the synthetic breast CT image

Kim,

Baek

2023

Phys. Med. Biol.

View full text Add to dashboard Cite

Objective: This paper proposes a new objective function to improve the quality of synthesized breast CT images generated by the GAN and compares the GAN performances for transfer learning datasets from different image domains.Approach: The proposed objective function, named beta loss function, is based on the fact that X-ray-based breast images follows the power-law spectrum. Accordingly, the exponent of the power-law spectrum (beta value) for breast CT images is approximately two. The beta loss function is defined in terms of L1 distance between the beta value of synthetic images and validation samples. To compare the GAN performances for transfer learning datasets from different image domains, ImageNet and anatomical noise images are used in the transfer learning dataset. We employee styleGAN2 as the backbone network and added the proposed beta loss function. The patient-derived breast CT dataset is used as the training and validation dataset; 7355 and 212 images are used for network training and validation, respectively. We use the beta value evaluation and Fréchet inception distance (FID) score for quantitative evaluation. Main results: For qualitative assessment, we attempt to replicate the images from the validation dataset using the trained GAN. Our results show that the proposed beta loss function achieves a more similar beta value to real images and lower FID score. Moreover, we observe that the GAN pretrained by anatomical noise images achieves better equality than ImageNet for beta value evaluation and FID score. Finally, the beta loss function with anatomical noise as the transfer learning dataset achieves the lowest FID score. Significance: Overall, the GAN using the proposed beta loss function with anatomical noise images as the transfer learning dataset provides the lowest FID score among all tested cases. Hence, this work has implications for developing GAN-based breast image synthesis methods for medical imaging applications.

show abstract

Section: Discussionmentioning

confidence: 99%

Power-law spectrum-based objective function to train a generative adversarial network with transfer learning for the synthetic breast CT image

Kim,

Baek

2023

Phys. Med. Biol.

View full text Add to dashboard Cite

show abstract

“…Several articles in the literature confirm that machine learning and in particular the deep learning (DL) is significantly performing in biomedical imaging processing 22 . In MRI framework, DL algorithms find many applications in several contexts like detection, segmentation, relaxometry or image translation 23–26 . In this manuscript, a deep learning algorithm is proposed, able to analyze MRI voxels composition in order to distinguish between homogenous and heterogeneous voxels.…”

Section: Methodsmentioning

confidence: 99%

“…22 In MRI framework, DL algorithms find many applications in several contexts like detection, segmentation, relaxometry or image translation. [23][24][25][26] In this manuscript, a deep learning algorithm is proposed, able to analyze MRI voxels composition in order to distinguish between homogenous and heterogeneous voxels. In particular, a supervised 27 machine learning approach is exploited through the use of a neural network (NN) algorithm trained on a labeled dataset.…”

Section: Methodsmentioning

confidence: 99%

Intra voxel analysis in magnetic resonance imaging via deep learning

Autorino,

Franceschini,

Ambrosanio

et al. 2023

Int J Imaging Syst Tech

View full text Add to dashboard Cite

Magnetic Resonance Imaging (MRI) is a useful diagnostic method for producing anatomical images of the body. It differentiates tissues thanks to the measure of their proton density and relaxation times and . Since pathological tissues often present altered , and with respect to the physiological ones, MRI is largely used for the detection of large number of conditions. In addition, MRI scan presents an effective resolution able to detect very small anatomical elements, making the imaging system well suitable in contexts like prevention and early diagnosis. Nevertheless, system resolution imposes pathological volume to be larger than the voxel dimension. Since in some pathologies and conditions it could be very helpful to find tissues even smaller than the voxel dimension, this manuscript proposes an algorithm able to analyze voxel content and detect which one presents more than one tissue in its volume (i.e., which one is heterogeneous). More in detail, a machine learning algorithm is proposed, able to highlight, in the MR image, which pixel corresponds to an heterogeneous voxel. Method shows to be promising in combining good results and near real‐time processing in both simulated and real scenario.

show abstract

“…Dalmaz et al [24] presented a new approach dependent upon adversarial diffusion modeling, SynDiff, to enhance the efficiency of medical image translation. For capturing a direct connection of the image distribution, SynDiff leverages a conditional diffusion procedure which gradually maps the noise and source image onto the target image.…”

Section: Related Workmentioning

confidence: 99%

Dung Beetle Optimization with Deep Feature Fusion Model for Lung Cancer Detection and Classification

Alamgeer

Alruwais

Alshahrani

et al. 2023

Cancers

View full text Add to dashboard Cite

Lung cancer is the main cause of cancer deaths all over the world. An important reason for these deaths was late analysis and worse prediction. With the accelerated improvement of deep learning (DL) approaches, DL can be effectively and widely executed for several real-world applications in healthcare systems, like medical image interpretation and disease analysis. Medical imaging devices can be vital in primary-stage lung tumor analysis and the observation of lung tumors from the treatment. Many medical imaging modalities like computed tomography (CT), chest X-ray (CXR), molecular imaging, magnetic resonance imaging (MRI), and positron emission tomography (PET) systems are widely analyzed for lung cancer detection. This article presents a new dung beetle optimization modified deep feature fusion model for lung cancer detection and classification (DBOMDFF-LCC) technique. The presented DBOMDFF-LCC technique mainly depends upon the feature fusion and hyperparameter tuning process. To accomplish this, the DBOMDFF-LCC technique uses a feature fusion process comprising three DL models, namely residual network (ResNet), densely connected network (DenseNet), and Inception-ResNet-v2. Furthermore, the DBO approach was employed for the optimum hyperparameter selection of three DL approaches. For lung cancer detection purposes, the DBOMDFF-LCC system utilizes a long short-term memory (LSTM) approach. The simulation result analysis of the DBOMDFF-LCC technique of the medical dataset is investigated using different evaluation metrics. The extensive comparative results highlighted the betterment of the DBOMDFF-LCC technique of lung cancer classification.

show abstract

Unsupervised Medical Image Translation With Adversarial Diffusion Models

Cited by 129 publications

References 64 publications

Power-law spectrum-based objective function to train a generative adversarial network with transfer learning for the synthetic breast CT image

Power-law spectrum-based objective function to train a generative adversarial network with transfer learning for the synthetic breast CT image

Intra voxel analysis in magnetic resonance imaging via deep learning

Dung Beetle Optimization with Deep Feature Fusion Model for Lung Cancer Detection and Classification

Contact Info

Product

Resources

About