Meeting Banner
Abstract #1775

MRI Based Artificial Neural Network Model Used in Prostate Cancer Detection

Chengyan Wang1, Juan Hu2, He Wang2, Hui Zhang1, Rui Wang2, Wenchao Cai2, Wei Wang2, Xiaoying Wang, 12, Jue Zhang3, 4, Jing Fang, 13

1Academy for Advanced Interdisciplinary Studies, Peking University, Beijing, China; 2Department of Radiology, Peking University First Hospital, Beijing, China; 3College of Engineering, Peking University, Beijing, China; 4Academy for Advanced Interdisciplinary Studies, Peking University , Beijing, China

In the present study, we propose a multi-layer artificial neural network (ANN) model integrating MR images (T2 weighted imagesdiffusion weighted images, dynamic contrast enhanced images and MR spectroscopy) and clinical examination (Total PSA value, free/total PSA ratio and age) for prostate cancer detection. The new model is able to provide high accuracy in detection of prostate cancer through proper training. Significant improvement in the AUC can be produced when compared to clinical-only model, and this demonstrates the capacity of MRI for computer-aided prostate cancer identification.

Abstract PDF Poster PDF

Keywords

able academy accuracy achieved advanced aided among antigen applicable artificial back belonging better biopsy black blue bray cancer caner carried changing china chosen circles classifier clinical college combining common computer condition confirmed connection containing contrast cross curve curves database detected detection devised diagnosed diagram diffusion dynamic early effectiveness either engineering enhanced examination examinations exhibited fang feasibility features fifth fold free frequently good green hidden highly hospital identification important included increasingly indicate indicators input inputs integrated integrating interconnected interdisciplinary layer layers learning least marker materials medical model models network neural neurons output overall patient patients performance plays potential presently probability produces propagation proper proposed prostate purple quantitative radiology raised random randomly recruited remaining respectively richly role rule samples scanners schematic selected serial serum sets significance significantly source spectroscopy stage staging statistical statistically studies suggested summary surgery suspected system systems table training tumor urologist utilizing validation weighed widely years