Computer Aided Detection for Radiologic Images

Clinical Center

Investigators

Linked publications, trials & patents

Abstract

The purpose of this project is to develop computer-aided diagnosis/detection (CAD) for a wide variety of radiologic images and disease types. This project uses existing NIH radiology images. We are developing techniques for segmentation of abdominal CT images to accurately locate the boundaries of the major abdominal organs such as the liver, spleen, kidneys, muscles and pancreas. We made further progress on this project, providing accurate localization and measurement of diseases such as abdominal atherosclerotic plaques. We made further progress on a project to develop computer-aided assessment of body composition on CT scans. We are developing convolutional neural networks based methods ("deep learning") on big data to train computers to detect diseases on radiology images like X-Ray, CT and MRI scans. In FY 2023, we made several scientific advances. These included (1) improved universal lesion detection from CT to multiparametric MRI potentially improving diagnosis for a wide range of diseases, (2) showing that sarcopenia used in routine pre-liver transplant abdominal CT was the only factor significantly associated with post-LT mortality, (3) improved automated detection of small bowel carcinoid tumors on CT, and (4) automated longitudinal assessment of volumetric renal stone burden on CT scans.