Integrating Clinical Knowledge in a Thyroid Nodule Classification Model Based on

Deep neural network models are currently facing problems due the weak interpretability of predicted results. One solution would involve integrating the clinical knowledge and classifiers. In this study, we utilized mature clinical experiences from TI-RADS, and also trained a multi-label cost-sensitive classification network. The results show that the end-to-end classification model based on cost-sensitive loss function is a more effective way to integrate clinical knowledge into the deep neural network.