31P-MRS data analysis of liver based on self-organizing map neural networks

Objective: Discussion based on neural networks in the ³¹P MR spectroscopy to distinguish hepatocellular carcinoma, normal liver and cirrhosis in value. Methods: Using self-organizing map neural network (SOM) analyse 66 data of ³¹P MRS, including hepatocellular carcinoma (13 samples), normal liver (16 samples) and liver cirrhosis (37 samples). Results: ³¹P MRS can be used for the diagnosis and differential diagnosis between hepatocellular carcinoma and liver cirrhosis nodules. The four experiments show that neural network model based on the ³¹P MR spectroscopy data analysis may increase diagnostic accuracy rate of hepatocellular carcinoma from 85.4% to 92.31%. Conclusion: ³¹P MRS data analysis based on neural network model provides a valuable diagnostic means of of hepatocellular carcinoma in vivo.