Studying the Applicability of Generative Adversarial Networks on HEp-2 Cell Image Augmentation

Abstract

The Anti-Nuclear Antibodies (ANAs) testing is the primary serological diagnosis screening test for autoimmune diseases. ANAs testing is conducted mainly by the Indirect Immunofluorescence (IIF) on Human Epithelial cell-substrate (HEp-2) protocol. However, due to its high variability, humansubjectivity, and low throughput, there is an insistent need to develop an efficient Computer-Aided Diagnosis system (CADs) to automate this protocol. Many recently proposed Convolutional Neural Networks (CNNs) demonstrated promising results in HEp-2 cell image classification, which is the main task of the HE-p2 IIF protocol. However, the lack of large labeled datasets is still the main challenge in this field. This work provides a detailed study of the applicability of using generative adversarial networks (GANs) algorithms as an augmentation method. Different types of GANs were employed to synthesize HEp-2 cell images to address the data scarcity problem. For systematic comparison, empirical quantitative metrics were implemented to evaluate different GAN models' performance of learning the real data representations. The results of this work showed that though the high visual similarity with the real images, GANs' capacity to generate diverse data is still limited. This deficiency in the generated data diversity is found to be of a crucial impact when used as a standalone method for augmentation. However, combining limited-size GANsgenerated data with classic augmentation improves the classification accuracy across different variants of CNNs. Our results demonstrated a competitive performance for the overall classification accuracy and the mean class accuracy of the HEp-2 cell image classification task.INDEX TERMS Computer-aided diagnosis systems (CADs), Convolutional Neural Networks (CNNs), Data augmentation, Data diversity, Evaluation metrics, Generative adversarial networks (GANs), HEp-2 cell image classification.

Deep Active Learning for Automatic Mitotic Cell Detection on HEp-2 Specimen Medical Images

“…For instance, many initial studies attempted to approach this task using conventional learning methods, such as the works proposed in [ 12 , 18 , 19 , 20 , 21 ]. Alternatively, the later works proposed CNN-based approaches, which demonstrated superior performance over the conventional handcrafted-based learning methods [ 6 , 22 , 23 , 24 , 25 , 26 , 27 , 28 ].…”

“…Later, Majtner et al [ 28 ] employed a separated DCGAN model to synthesize new images of each HEp-2 interphase type class for augmentation purposes. In more recent work, Anaam et al [ 6 ] studied the effectiveness of using different GAN models to generate new HEp-2 cell images for boosting the CNN classification performances. However, for the specific task of HEp-2 mitotic cell classification, the works proposed by Gupta et al [ 37 ] and Anaam et al [ 38 ], introduced in Section 2.1.1 , are the only published studies.…”

“…However, a small portion of this interest is directed to approaching the task of mitotic cell identification even though it is a crucial step in the IIF HEp-2 diagnostic procedure, and hence, for developing a feasible HEp-2 CAD system [ 4 ]. Recently, the CAD system based on deep learning has gotten much attention for automating the prediction process with a massive number of extracted deep features, improving detection performance, and reducing user intervention during the training process [ 6 , 7 ].…”

Generative adversarial networks in cell microscopy for image augmentation. A systematic review

“…Five studies implemented auxiliary losses, eleven used the GAN architectures as intermediate tasks (to boost classification, segmentation, and object detection), and eleven were implementations of already existing models without any modifications. There are studies using PGGAN [41] [42], sinGAN [43], StyleGAN [44] immunofluorescence images [50].…”

MBAHIL: Design of a Multimodal Hybrid Bioinspired Model for Augmentation of Hyperspectral Imagery via Iterative Learning for Continuous Efficiency Enhancements

“…For instance, works in [15], [16], [17], and [18] propose the use of a Generative Adversarial Network (GAN), YOLOv5s, a multi-semantic global channel, and spatial joint attention module (MsGCS) for the estimation of augmented image sets for different application scenarios. These scenarios are extended in [19], [20], which discuss using Soft Augmentation-Based Siamese CNN (SAB SCNN) and different GANs for hyperspectral image sets. These models are highly scalable but showcase higher complexity, which limits their speed performance levels.…”