甲状腺结节分析数据

1：甲状腺结节的评估与筛查：辅助评估甲状腺结节的良恶性。
2：基于甲状腺结节分析数据，可以开发人工智能（AI）甄别系统，自动分割和诊断甲状腺结节良恶性，辅助低年资医师更准确鉴别甲状腺结节良恶性。1.数据采集：采集病例的监测设备、型号、患者性别、年龄、等信息（不涉及能够识别自然人个人身份的信息），人工标注录入病例分级（类）、mask_勾画轮廓。
2.算法获得诊断结果，使用EfficientNet-B6作为基础网络，在标注的大数据基础上，使用“小尺寸训练、大尺寸微调”的训练方法，解决使用在线数据扩增方法引起的训练和测试时物体分辨率不一致问题，按照有监督学习锐度感知最小化方式优化模型参数，同时减小损失值和损失锐度，提高模型的泛化能力。使用焦点损失函数，解决结节识别难易程度不均衡问题，提高困难结节的学习权重。使用多模型交叉融合方法，将预测平均值作为最终输出的诊断概率值，在调优集上选取合适阈值确定软件产品的灵敏度和特异度。

1. Thyroid Nodule Evaluation and Screening: Assisting in the assessment of benign and malignant thyroid nodules.
2. Based on the analysis data of thyroid nodules, an artificial intelligence (AI) screening system can be developed to automatically segment thyroid nodules and diagnose their benign and malignant nature, assisting junior physicians in more accurately differentiating between benign and malignant thyroid nodules.
1. Data Collection: Collect information such as monitoring equipment, model, patient gender, and age of the cases (no information that can identify the personal identity of natural persons is included). Manually annotate and input the case grade (category) and delineate the mask contour.
2. The algorithm generates diagnostic results. EfficientNet-B6 is used as the basic network. Based on the labeled large-scale dataset, the training method of "small-size training, large-size fine-tuning" is adopted to solve the problem of inconsistent object resolution between training and testing caused by online data augmentation. The model parameters are optimized via supervised learning-based Sharpness-Aware Minimization, which simultaneously reduces both the loss value and loss sharpness to enhance the model's generalization ability. The focal loss function is utilized to address the class imbalance issue in nodule recognition, thereby increasing the learning weight for challenging nodules. A multi-model cross-fusion approach is employed, taking the average of the prediction results as the final output diagnostic probability value, and an appropriate threshold is selected on the tuning set to determine the sensitivity and specificity of the software product.

搜集汇总

数据集介绍

特点

甲状腺结节分析数据集由浙江德尚韵兴医疗科技有限公司提供，包含806条记录，每月更新。数据集用于甲状腺结节的评估与筛查，支持开发AI甄别系统，辅助医师准确鉴别甲状腺结节的良恶性。数据包括病例的设备信息、患者基本信息、病例分级及诊断结果等，采用先进的算法规则进行数据处理和模型优化。