eISSN: 2373-6372

Gastroenterology & Hepatology: Open Access

Correspondence:

Received: January 01, 1970 | Published: ,

Citation: DOI:

Download PDF

Editorial

In this editorial, we compared conventional statistical analysis of liver disease data with neural network analysis to predict which risk factors are more important in the prediction of liver diseases. Usually, liver function tests are used to assess liver diseases. These risk factors include, but not restricted to, ALT, AST, total bilirubin and direct bilirubin, total proteins, alkaline phosphatase, and albumin. Some demographical characteristics are also included such as age and gender.

Conventional statistical analysis modes include descriptive analysis such as frequency, percentages, mean, and standard deviations. The relationships between study variables can be assessed using the difference in means such as T test, and One Way ANOVA. Correlations and regressions can also be used to give in depth sights of the liver diseases. The output of these analyses is well known.

With the development of science, machine learning or artificial intelligence has been created to simulate the way of brain activities in thinking. Much progress has been made such as the prediction mode. In this mode of analysis, datasets for various diseases have been created including diabetes, heart, and liver. In general, datasets are designed to have output of two classifications as either diseased or not diseased. The datasets have the data of risk factors, or predictors of liver disease. The artificial intelligence part of this data implies thinking as human brain does. There are three layers in the artificial intelligence, input layer (predictors), hidden layers, and output layer. The system will run many times to give a prediction of which predictors are most important. In the following example, a comparison between conventional and artificial network analysis was made for the same liver disease dataset. Table 1 showed the general characteristics of study variables including frequency, percentage for categorized variables, and mean and standard deviation for non-categorized variables. Table 2 showed the relationships between study variables using independent T test. This test can provide important information including mean, standard deviation, and significance. When significance is equal, it is difficult to determine the relative importance of each predictor.

Variable	Description
Age (M±SD) years	44.75 (16.19)
Gender (N, %): - Males - Females	441 (75.6%) 142 (24.4%)
Total bilirubin (M±SD) mg/dl	3.30±6.21
Direct bilirubin (M±SD) mg/dl	1.49±2.81
Alkaline phosphatase (M±SD) IU/L	290.58±242.4
ALT (M±SD) IU/L	80.71±182.62
AST (M±SD) IU/L	109.91±288.92
Total protein (M±SD) g/dl	6.48±1.09
Albumin (M±SD) g/dl	3.14±0.80
Albumin/globulin (%)	0.974±0.32
Health status (N, %): - Diseased - Normal	416 (71.4%) 167 (28.6%)

Table 1 General description of study variables

Variable	Dataset	N	Mean	Std. Deviation	P value
Age	Disease	416	46.15	15.65	0.001
Age	Normal	167	41.24	16.99	0.001
Total bilirubin	Disease	416	4.16	7.14	0.000
Total bilirubin	Normal	167	1.142	1.00	0.000
Direct bilirubin	Disease	416	1.92	3.20	0.000
Direct bilirubin	Normal	167	0.39	0.52	0.000
Alkaline phosphatase	Disease	416	319.00	268.30	0.000
Alkaline phosphatase	Normal	167	219.75	140.98	0.000
ALT	Disease	416	99.60	212.76	0.000
ALT	Normal	167	33.65	25.06	0.000
AST	Disease	416	137.69	337.38	0.000
AST	Normal	167	40.68	36.41	0.000
Total protein	Disease	416	6.45	1.09	0.399
Total protein	Normal	167	6.54	1.06	0.399
Albumin	Disease	416	3.06	0.78	0.000
Albumin	Normal	167	3.34	0.78	0.000
Albumin/globulin_ratio	Disease	414	0.91	0.33	0.000
Albumin/globulin_ratio	Normal	165	1.03	0.29	0.000

Table 2 The relationships between study variables using independent T test

Using artificial intelligence gives various figures and tables of which Table 3 showed the importance of independent variables. In this table, the importance is given as proportions and percentages. In Figure 1, normalized importance is given according to their importance.

	Importance	Normalized importance
Age	.063	33.9%
Total bilirubin	.099	53.6%
Direct bilirubin	.121	65.7%
Alkaline phosphatase	.053	28.9%
AST	.169	91.6%
ALT	.185	100.0%
Total protein	.109	58.9%
Albumin	.144	78.1%
Albumin/globulin ratio	.056	30.4%

Table 3 The importance of independent variables

Figure 1 Normalized importance of covariates on liver disease.

Conclusion

Although conventional statistical analysis of liver data can provide important information about liver state conditions, appropriate prediction of health and disease requires neural network analysis.