Generative Adversarial Networks in Medical Image augmentation: A review

Synthetic 18F-FDG PET Image Generation Using a Combination of Biomathematical Modeling and Machine Learning

“…However, the effectiveness of these techniques on medical image analysis with complex imaging textures is not as useful as the conventional image datasets because the medical image patterns might be changed by applying basic geometric and deformable augmentation techniques such as translation and rotation. Synthetic data generation is another type of data augmentation approach that is able to programmatically learn the representations of images, produce realistic images to build the model’s generalizability, and decrease overfitting during the training process [ 7 , 8 ]. Such synthetic generated datasets are extremely useful for medical image analysis.…”

“…The GAN has two deep architecture functions, generator and discriminator, and it is trained in an adversarial fashion in which the generator produces the fake samples, and the discriminator iteratively trains to distinguish between fake and real samples. Several GAN-based models have been developed for the different application tasks of image segmentation, detection, classification, registration, super-resolution, and denoising [ 7 , 8 ]. Nie et al [ 10 ] proposed an adversarial strategy to train a fully convolutional network, aiming to generate synthetic pelvic computed tomography (CT) images given input magnetic resonance (MR) images.…”

HADCNet: Automatic segmentation of COVID-19 infection based on a hybrid attention dense connected network with dilated convolution

“…We design the spatial attention with dilated convolution, combining the 3×3 convolution kernel and dilated convolutions with expansion rates of [ 1 , 2 , 5 ], replacing the dilated convolution with an expansion rate of 1 with a densely connected block, as shown in Equation (1) , where and represent dilated convolutions with expansion rates of 2 and 5, respectively. is a densely connected block, which includes 3×3 three convolution layers, represents the Conv3×3-BN-ReLU structure, and the value after combining the three convolution layers is the final output of the dense connection.…”

“…The segmentation of lesions from medical images can provide doctors with critical information for diagnosing and quantifying diseases. Traditional medical image segmentation algorithms typically extract features manually using physical information such as the texture, structure and location of the image, with a reliance on extensive pre-processing operations and an experienced manual design process [ 5 ]. Segmentation methods based on machine learning achieve the segmentation goal by focusing on algorithms and predicting personalized features.…”

Multilevel threshold image segmentation for COVID-19 chest radiography: A framework using horizontal and vertical multiverse optimization

“…There are many methods to process images efficiently in the computer field [ [7] , [8] , [9] ], among which image segmentation is an integral part of image processing and a hot topic of research nowadays [ [10] , [11] , [12] ]. In the past few years, more and more image segmentation techniques have been proposed, such as multilevel threshold image segmentation (MTIS) [ 13 ], deep learning-based image segmentation [ 14 ], hierarchical clustering-based image segmentation [ 15 ], wavelet transform-based image segmentation [ 16 ], and others.…”