基于超声RF信号的甲状腺结节智能诊断方法*

由于人口老龄化的原因,甲状腺癌的发病率增长率在所有癌症中是最为显著的。因此,对存在癌变可能的甲状腺结节进行预检查显得尤为重要,而超声智能诊断系统在甲状腺结节早期筛查方面已展现出巨大的应用前景。该文的工作旨在提出一种基于超声原始射频信号的组织参数定征和人工神经网络相结合的甲状腺结节智能诊断方法,以提高临床超声诊察效率及准确性。为达成上述目的,该文使用滑动窗口图像分析方法和多兴趣区覆盖的方法提取组织定征参数作为特征,使用人工神经网络进行良恶性分类,并对可能影响分类准确性的相关因素进行参数相关性分析。结果显示,基于临床样本,该文提出的智能诊断方法可达到93.2%敏感度、94.0%特异性和93.5%准确率。该方法一定程度上克服了传统方法无法充分利用图像局部细节信息的不足,有效提高了诊察效率和准确性;另一方面,与深度神经网络相比,本方法对计算资源和样本量的需求较少。因此有望在该文研究基础上最终建立一套可实际用于甲状腺结节的预筛查的临床智能诊断系统。

Computer-aided diagnosis of thyroid nodules based on ultrasound RF signal

Due to the aging of the population, the incidence of thyroid cancer dramatically increases in recent years. Therefore, pre-examination of thyroid nodules is particularly important. Since ultrasound intelligent diagnosis system has shown great application prospects in the early screening of thyroid nodules, this paper aims to propose a new intelligent diagnosis method of thyroid nodules, which combines tissue parameter characterization and artificial neural network based on ultrasound RF signals, to improve the efficiency and accuracy of clinical ultrasound diagnosis. To achieve this purpose, we use sliding window image analysis method and multi-ROI coverage method to extract tissue characteristic parameters as features, then use artificial neural network as classifier, and analyze the correlation of related factors that may affect the accuracy of classification. The results showed that the application of the current method can achieve a sensitivity of 93.2%, a specificity of 94.0% and an accuracy of 93.5%. To a certain extent, this method overcomes the inability of traditional methods to make full use of the local details, thereby effectively improving the efficiency and accuracy of diagnosis. On the other hand, compared with Deep Neural Networks, it has less demand for computational resources and sample size. It is expected to finally establish a clinical intelligent diagnosis system that can be practically used for pre-screening of thyroid nodules.