Credence-Net: a semi-supervised deep learning approach for medical images

Deep learning uses a large-scale labelled dataset to ensure a high degree of accuracy. This technology is increasingly data-driven in medicine and biology imaging, and labelled data is more difficult and expensive to retrieve. Various studies are being conducted on semi-supervised deep learning models (SSDLM) and self-supervised deep learning. In order to increase the quantity of labelled data necessary for deep learning, researchers are increasingly looking at SSDLM and its applications. The motivation for the proposed Credence-Net is similar to how physicians handle uncertain or questionable instances in reality, based on their colleague's or senior's consultation. Proposed model Credence-Net has attained the best accuracy and specificity, sensitivity, precision, Matthews correlation coefficient, false discovery rate, false-positive rate, f1 score, negative predictive value, and false-negative rate 91.834%, 85.268%, 97.008%, 89.356%, 83.648%, 10.644%, 14.732%, 93.016%, 95.696%, and 2.992% for unseen dataset respectively. This research work leads to a more accurate and efficient SSDLM.