Parathyroid Gland Detection in Nuclear Medecine

This repository contains different codes of a generative model NM-USNet that is a combination of a Siamese Network and a U-Net model for parathyroid glands detection through automatic image subtraction using 123-I and 99m-Tc images. The combination of the two components is monitored with an optimized loss function that includes reconstruction error (MSE of the U-NET) and similarity computed by the SN, with dynamic weighting adjustment methodology.

本仓库收录了生成式模型NM-USNet的各类代码，该模型将孪生神经网络（Siamese Network）与U-Net架构相结合，用于通过123-I与99m-Tc影像实现自动图像减影的甲状旁腺检测任务。该双组件的结合通过优化后的损失函数进行监控，该损失函数包含重建误差（即U-Net的均方误差MSE）以及由孪生神经网络计算得到的相似度，并采用动态权重调整方法。