Segmentation method for breast tumor diagnosis based on Artificial Neural Network algorithm applied to dynamic 18F-FDG PET images

Abstract

To establish an accurate and automatic image segmentation method for extracting the tumor from the healthy tissue using the dedicated 18F-FDG Mammography with Molecular Imaging (MAMMI) Positron Emission Tomography (PET) dynamic images of breast in vivo, the paper presents a novel method using Artificial Neural Network (ANN) combined with time activity curves (TAC) of each voxel as input vector. TACs voxel by voxel were obtained from PET images with average filtering and least-squares fitting algorithm to improve the signal noise ratio (SNR). The data was then normalized and constructed as input feature vectors of the ANN network to train or segment the tumor regions. The initial pilot validation with 2 patient's data of the proposed method using FDG has shown promising results.