Alcoholic liver diseases are associated with a high incidence of death globally. But there is still a lack of computational diagnosis and classification of alcoholic hepatitis. Classification of alcoholic liver diseases is important as the different stages of the disease demands significantly different therapeutic strategies. 

A study utilized general regression neural network (GRNN) model with high-performance classification ability for diagnosing and classifying alcohol hepatitis. Tenfold cross-validation was utilized to demonstrate the error rate of networks. 

The results demonstrated that 80.91% accuracy of the back diagnosis in 110 patients. 81.82% accuracy of predicting-diagnosis in 11 patients referring to the clinical diagnosis made by a group of experts. 

The study utilized the architecture of the neural network, which is the multilayered feed forward network architecture with 7 input variables and 3 output indicators. The input variables were determined by 7 liver function tests while the output indicators were represented as the classification of the disease development stages. This GRNN model accurately staged alcoholic liver disease. 

Since the changes of these liver function tests were not disease-specific, they only reflected the degree of liver injury and substitute disease development stages. Thus these variables could not distinguish hepatitis caused by different etiologies.

GRNN can diagnose ALD with an accuracy of around 80.91%–81.82%. They also reduce the diagnosis time and improve the diagnosis accuracy. 

Thus the study suggested utilizing the liver function tests as the input layer variables of the GRNN model for accurately diagnosing and classifying alcoholic liver diseases. 

Source: Naiping Li, Yongfang Jiang, Jin Ma, Bo He, Wei Tang, Mei Li, Qing Huang, Ting Yuan, "Computational Classification and Diagnosis of Alcoholic Liver Diseases Using General Regression Neural Network", Mathematical Problems in Engineering, vol. 2014, Article ID 524621, 10 pages, 2014. https://doi.org/10.1155/2014/524621