基于反向传输神经网络的肝脏^(31)P磁共振波谱分析

^(31)P-MRS data analysis of liver based on back-propagation neural networks

目的探讨基于神经网络的31P磁共振波谱(31P-MRS)辨别肝硬化、肝细胞癌(HCC)和正常肝组织的价值。方法运用反向传输神经网络分析66个31P-MRS样本数据,其中包括37个肝硬化结节样本、13个HCC样本和16个正常肝脏样本。结果经交叉验证实验证明,基于神经网络模型的31P-MR波谱数据分析可以将肝细胞癌的诊断正确率从85.47%提高到92.31%。结论基于神经网络模型的31P-MRS波谱数据分析可以用于HCC与肝硬化结节的诊断和鉴别诊断。

Objective To explore the value of distinguishment of hepatocellular carcinoma （HCC）, cirrhosis nodules and normal liver based on neural networks in the ^31P-MR spectroscopy. Methods A total of 66 data of ^31P-MRS were analysed using back-propagation neural network, including 37 samples of liver cirrhosis, 13 samples of HCC and 16 samples of normal liver. Results The cross-valiation experiments showed that diagnostic accuracy rate of HCC increased from 85.47% to 92.31% with neural network model based on the ^31P-MR spectroscopy data analysis. Conclusion ^31P-MRS data analysis based on neural network model provides a valuable diagnostic tool of HCC in vivo.