Neural Network Application in Eco-Ultrasound Brightness Prediction Associated with Hepatic Diagnosis

Abstract

The imaging diagnostic is a useful tool which avoids further investigations. Usually, a lot of results are concurrent to diagnosis and a numerical tool with artificial intelligence could better use all these data to predict the stage of steatosis at the liver level. Our goal is to test a unified framework based on neural network application for image analysis in medical science. A number of 100 pacients are used for generating the data matrix in order to build a NEUROSTEATOSIS trained neural network. Ten variables are decided, seven are for input parameters and three for output parameters. The neural model establishes a correlation between the input and output parameters in order to predict interrogated values between the range of trained ones. The results were validated with some results based on another investigations applied to other pacients for whom the level of steatosis is known. This new tool could be further used for a preliminary estimation of the progress of steatosis in certain conditions.