Analysis of hepatic fibrosis risk factors using artificial neural networks

BOUHARATI KHAOULA

1,2*

BOUHARATI IMENE

3,4

, GUENIFI WAHIBA

1.Department of Epidemiology, Faculty of Medicine, Constantine University, ALGERIA

2.Laboratory of Health and Environment, UFAS Setif1 University, Setif, ALGERİA

3.Faculty of Medicine, Paris Sorbonne University, FRANCE

dysfunction, it switches to cirrhosis. The factors involved are sometimes ill-defined. However,

the most common are hepatitis B and C and alcohol abuse. The analysis of these factors is very

complex. Methods: This study proposes an artificial intelligence tool, in particular artificial

neural networks in data analysis. We consider risk factors as input variables to the system. We

consider the risk of fibrosis as an output variable. Conclusion: When the learning of the

network is carried out from the proper cases followed at our hospital service of Setif in Algeria,

the transfer function created is adjusted to its minimum of errors. It then becomes possible to

assign random values to the input of the system to read the risk of fibrosis at the output.

Keywords: Liver fibrosis, Risk factors, Intelligent analysis, ANN

collagen, which sometimes takes the form

of a dense network of fibers. At its

advanced stage, it transforms into nodular

cirrhosis, thus deforming the hepatic

vascular system [1], [2]. Often irreversible,

hepatic cirrhosis can lead to death.

They often relate known risk factors that

cause liver damage to gender, age, alcohol

abuse, diabetes and mainly viral load in

hepatitis B or C or [3 ].

hyperlipidemia is also associated with

fibrosis [5-8]. The accumulation of fat in

the liver can determine the prevalence of

fibrosis [9]. The association of this disease

with coronary heart disease (CHD) is also

mentioned [10]. Given this complexity, a

classification of the severity of fibrosis is

established according to a score. A low

probability when the score is between 1 and

5. Medium probability is between -1.5 and

technique based on the principles of

artificial intelligence, mainly the

application of artificial neural networks.

The constructed artificial neural network

has three layers (input layer, hidden layer,

and output layer). This is to make the

correspondence between the two spaces

(inputs-outputs). Risk factors are

the network is optimized, it then becomes

possible to assign random values to the

input of the system to read the risk of

fibrosis at the output.

2. Material and method

Fibrosis is a multi-factorial disease. What

characterizes these factors is complexity

and uncertainty. The risk factors are

multiple. Some factors are known, some

other less so. Also, the weight of each

mathematical techniques is very difficult

and even impossible. This study addresses

the analysis of these risk factors by artificial

neural networks. This technique makes it

possible to deal with the complexity and

multitude of factors involved.

The factors analyzed in this study are

limited only to the three major factors,

which are hepatitis B and C and excessive

alcohol consumption.

initial data is stored, processed and returned

at the output [20]. The major advantage of

artificial neural networks is learning. By

introducing values at the input and the

result at the output, a transfer function is

created [21-23]. At each value entered, the

function is readjusted so that it remains

appropriate. This is done just by variations

[25].

The network applied to this analysis is a

multi-layer network (Figure 1). It took risk

factors for hepatic fibrosis as input

variables to the system. These factors

(hepatitis B, hepatitis C and the age of the

patients). The alcohol abuse factor is not

supported because this factor did not arise

during our follow-up. The output variable

of the system expresses damage by fibrosis.

Mathematically, this function can be

represented by: df=fhb,hc,ap

blood. These diseases are characterized by

inflammation of the liver. The infection

often remains mild. However, sometimes

(about 10%), it develops into a chronic

infection and can even develop into

cirrhosis or even liver cancer. WHO

statistics, reports that the approximately

257 million infected cases are in a chronic

stage [12]. Viral hepatocyte infections

progress to acute hepatitis and even induce

hepatocellular carcinoma or ultimately

cirrhosis. Regardless of the existence of a

supposedly 95% effective vaccine,

hepatocellular carcinomas rank second

The severity and the risk of liver damage

is determined by the volume, frequency and

duration of alcohol consumption. While the

excessive alcohol consumption is an

aggravating factor for hepatitis C. Its effect

will be much more pronounced, especially

when it is combined with advanced age or

other factors such as obesity or

immunodeficiency [15-19].

Each variable is assigned numerically:

Hepatitis B: 1 (positive); 2 (negative)

(Figure 1). From the real values recorded, a

mapping between each input variable and

its corresponding output variable

The cases analyzed are 80 cases. Half is

taken for network learning. The adjustment

of the function is done after 100 iterations.

The variables specific to each newly

introduced patient are combined with the

at the output (Figure 3). The result obtained

at the output is the aggregation of all the

inputs. In this way, it will be as precise as

possible.

For this reason, classical mathematical

methods are unsuitable for analyzing this

kind of situation. Often, it is a matter of

applying statistical methods. These methods

arrive at imprecise results and remain in the

realm of the probable. The proposed

technique applies a domain of artificial

neural networks. These networks have the

assigning the result to it at the output. A

transfer function that links the inputs to the

output is created. With each new case, the

function is adjusted. In the case of the

study, three input variables are taken into

correspondence between these variables and

the degree of attack by hepatic fibrosis is

done. After learning, the result will be as

accurate as possible in terms of reading this

degree later. It should be noted that the

network remains extensible to other factors

that are not taken into account in this study.

[10]. Adams LA, Lymp JF, St Sauver J,

Sanderson SO, Lindor KD, Feldstein A,

Angulo P. The natural history of nonalcoholic

fatty liver disease: a population-based cohort

study. Gastroenterology 2005; 129: 113-121.

DOI: 10.1053/ j.gastro.2005.04.014]

[11]. Sombat Treeprasertsuk, Einar Björnsson,

Felicity Enders, Sompongse Suwanwalaikorn,

Keith D Lindor. NAFLD fibrosis score: A

prognostic predictor for mortality and liver

complications among NAFLD patients. World

Liver. EASL clinical practice guidelines:

infection. J Hepatol 2012; 57 : 167-85.

[14]. Ferlay J, Soerjomataram I, Dikshit R, et al.

Cancer incidence and mortality worldwide:

Sources, methods and major patterns in

GLOBOCAN 2012: Globocan 2012. Int J

Cancer 2015 ; 136 : E359-86.

[15]. Thomas DL, Seeff LB. Natural history of

hepatitis C. Clin Liver Dis 2005. 9: 383,98,

DOI:https://doi.org/10.1016/j.cld.2005.05.003.

Interval: Necrobiome Analysis Using Artificial

Neural Networks. Computational Biology and

Bioinformatics. 2017. 5(6): 90-96.