基于自学习脉冲神经膜系统的早期肺癌分型智能诊断及其在临床中的应用研究

早期肺癌分型的精确诊断是采取合理治疗手段、降低肺癌死亡率的关键所在。受限于早期肺癌的隐匿性以及诊察手段的局限性，早期肺癌的人工诊断率很低。随着肺癌全球发病率逐年提升，肺癌诊疗数据大量积累。肺癌诊断逐渐从传统的人工诊断，转向大数据驱动的智能诊断。但“堆积如山”的肺癌数据中早期肺癌数据量极少且信息不完备，“小样本”成为深度学习等针对大规模数据的智能方法应用于早期肺癌分型诊断的天然壁垒。脉冲神经膜系统作为一类“轻量化”分布式并行计算模型，在处理“小样本”数据上展现出了优势。本项目拟针对早期肺癌数据的“小样本”特点，构建自学习脉冲神经膜系统对“小样本”数据进行自学习，进而发展早期肺癌分型诊断智能方法，并在两家合作的三甲医院进行临床应用示范。这方面的研究不仅在完善脉冲神经膜系统理论体系，以及在发展针对“小样本”数据自学习方法方面具有重要的理论价值，在早期肺癌分型诊断上也具有重要的临床应用价值。

Accurate type classification of early-stage lung cancer is the key to adopt reasonable treatment and reduce the mortality of lung cancer. Limited by the concealment of early-stage lung cancer and the diagnosis methods, the artificial diagnosis rate of early-stage lung cancer is very low. With the lung cancer increasing, the data of lung cancer has accumulated a lot. Lung cancer diagnosis has gradually shifted from manual diagnosis to intelligent diagnosis driven by large data. However, there are few data and incomplete information about early-stage lung cancer in the "mountain-like" lung cancer data. "Small sample" becomes a barrier to the application of deep learning intelligent methods, which are suitable for large-scale data. As a kind of "lightweight" and distributed parallel computing model, spiking neural P systems showed advantages in processing "small sample" data. According to the characteristics of "small sample" data, this project intends to construct a self-taught learning spiking neural P system, which has the ability of self-taught to "small sample" data. An intelligent method for the type classification of early-stage lung cancer will be developed, and will be demonstrated in two cooperative hospitals. The research has important values not only in improving the theoretical system, but also in developing self-taught learning methods for "small sample" data. The developed type classification method of early-stage lung cancer also has important clinical application value.