公开(公告)号：JPH09179843A

公开(公告)日：1997-07-11

申请号：JP26851596

申请日：1996-10-09

Applicant: ARCH DEV CORP

Inventor： UEI TSUAN ,  DOI KUNIO

IPC: A61B6/00 ,  G06F15/18 ,  G06N3/00 ,  G06N3/08 ,  G06Q50/00 ,  G06T1/40 ,  G06T7/00 ,  G06F19/00

Abstract: PROBLEM TO BE SOLVED: To provide the shift unchanged artificial neural network for detecting disease in a digital image. SOLUTION: This network includes an input layer and at least hidden layer and an output layer includes at least one neuron. Each hidden layer includes at least one hidden group. Each hidden group having at least one hidden unit includes at least one neuron. For learning using error back propagation algorithm so that the mean weight of a group becomes substantially zero, respective neurons need to be equal in connection weight between two groups in adjacent layers. A digital image of a patient is obtained to extract a suspicious area of interest and this area of interest is converted into numeric data, which are inputted to the neural network after the learning to detect a fine calcare part.