High‐resolution knee plain radiography image synthesis using style generative adversarial network adaptive discriminator augmentation

Abstract

In this retrospective study, 10,000 anteroposterior (AP) radiography of the knee from a single institution was used to create medical data set that are more balanced and cheaper to create. Two types of convolutional networks were used, deep convolutional GAN (DCGAN) and Style GAN Adaptive Discriminator Augmentation (StyleGAN2-ADA). To verify the quality of generated images from StyleGAN2-ADA compared to real ones, the Visual Turing test was conducted by two computer vision experts, two orthopedic surgeons, and a musculoskeletal radiologist. For quantitative analysis, the Fréchet inception distance (FID), and principal component analysis (PCA) were used. Generated images reproduced the features of osteophytes, joint space narrowing, and sclerosis. Classification accuracy of the experts was 34%, 43%, 44%, 57%, and 50%. FID between the generated images and real ones was 2.96, which is significantly smaller than another medical data set (BreCaHAD = 15.1). PCA showed that no significant difference existed between the PCs of the real and generated images (p > 0.05). At least 2000 images were required to make reliable images optimally. By performing PCA in latent space, we were able to control the desired PC that show a progression of arthritis. Using a GAN, we were able to generate knee X-ray images that accurately reflected the characteristics of the arthritis progression stage, which neither human experts nor artificial intelligence could discern apart from the real images. In summary, our research opens up the potential to adopt a generative model to synthesize realistic anonymous images that can also solve data scarcity and class inequalities.

A practical guide to the development and deployment of deep learning models for the orthopaedic surgeon: Part III, focus on registry creation, diagnosis, and data privacy

“…Every study reviewed above involves analysing the input image or text data and predicting certain outputs (i.e., image label, bounding box, segment of image or text), which encodes clinically meaningful information. In contrast to predictive deep learning models, generative AI, which is typically a deep learning model, reverses the process by synthesizing the original data conditioned on certain label information [1,19,38]. This synthetic data can be used to replace healthcare data of real patients for data transfer purposes as an alternative to deidentifying the data.…”

A practical guide to the development and deployment of deep learning models for the orthopedic surgeon: part II

“…Deep generative models have recently been used in other medical specialties to augment datasets [6] and resolve issues related to patient-privacy [20]. In orthopedics, one deep generative model was used to generate high-resolution knee radiographs that accurately relected the characteristics of arthritis progression through varying stages [1]. Ultimately, deciding which type of learning to utilize in a deep learning pipeline depends on the speciic purpose of the particular project.…”

Classifying E.coli concentration levels on multispectral fluorescence images with discrete wavelet transform, deep learning and support vector machine.

“…In the cases of data scarcity like ours, generative adversarial networks (GANs) are widely used (Ahn et al, 2023;Hermosilla et al, 2021;Yi et al, 2019;Zhang et al, 2018). An improved version of GAN known as Style generative adversarial networks-adaptive discriminator augmentation (StyleGAN2-ADA) has been shown to generate realistic synthetic images that can be used for generating larger datasets for training deep learning and machine learning models (Ahn et al, 2023). Based on their findings, we used StyleGAN2-ADA for generating larger datasets for training the VGG19 and SVM algorithms.…”