Semi-Supervised Medical Image Classification Combined with Unsupervised Deep Clustering

An effective way to improve the performance of deep neural networks in most computer vision tasks is quantity labeled data and quality labels. However, analysis processing medical images, high-quality annotation depends on experience professional knowledge experts, which makes it very difficult obtain a large number annotations. Therefore, we propose new semi-supervised framework for image classification. It combines classification with unsupervised clustering. Spreading label information unlabeled by alternately running two helps model extract semantic from data, prevents overfitting small amount data. Compared current methods, our enhances robustness reduces influence outliers. We conducted comparative experiment public benchmark dataset verify method. On ISIC 2018 Dataset, method surpasses other methods more than 0.85% AUC 1.08% Sensitivity. ICIAR BACH dataset, achieved 94.12% AUC, 77.92% F1-score, 77.69% Recall, 78.16% Precision. The error rate at least 1.76% lower that methods. result shows effectiveness