Semi-Supervised Robust One-Class Classification in RKHS for Abnormality Detection in Medical Images

Hybrid Graph Convolutional Network for Semi-Supervised Retinal Image Classification

“…Semi-supervised learning can exploit discriminative information in unannotated data to accelerate the supervised learning with limited labeled samples, and it can achieve robust models with insufficient annotations [15], which is more expensive in medical image annotation, and has excellent results compared to unsupervised methods [41]. To alleviate the annotation cost, several semi-supervised medical image classification models have been proposed [18], [22], [46], [49]. For example, Zhou et al [49] designed a jointly training model with semi-supervised framework to implement the DR grading and damage segmentation, guided by an attention module.…”

“…Xie et al [46] designed an adversarial learning mechanism under semi-supervised framework to conduct CT classification, which contains an adversarial auto-encoder R to implement the unsupervised self-expression, an identification network C trained by labeled data, and several trainable transforming layers that are in charge of transferring the image representations learned by R into the identification network C. Madani et al [22] utilized generative adversarial networks to leverage the imbalance of limited labeled samples and massive unannotated data under semi-supervised framework. Kumar et al [18] introduced robust statistical model to extend the multi-variable Gaussian generator into a scalable kernel Hilbert space under semi-supervised training, which can fully exploit the identical information in the limited annotated data.…”