Quality and Safety in Medical Imaging

Overview: I develop machine learning tools that automate time consuming and cumbersome task in healthcare quality assurance. Automated measures of quality offer major advantages over manual approaches to quality in terms of time, scale, and cost. This work will enable providers, researchers, and regulators to more effectively measure and monitor quality and value in healthcare.

Automated Protocol Selection


In diagnostic imaging, protocols are the precise set of instructions which define exactly how a set of medical images should be acquired. If acquired incorrectly, these images may be of little value to the radiologist interpreting the images and leave the patient without a critical diagnosis. We developed a machine learning algorithm to automate the process of protocol selection as well as prioritization of MRI brain examinations. This work will enable greater efficiency and personalization while reducing medical errors.

A Natural Language Processing Based Model to Automate MRI Brain Protocol Selection and Prioritization
A.D. Brown & T.R. Marotta. Academic Radiology 2016

CT Artifact Detection


The National Lung Screening Trial (NLST) found that people who got low-dose CT had a 20% lower chance of dying from lung cancer than those who got chest x-rays. Payers are now offering low dose CT for high risk patients. With more centers offering low dose CT screening exam, what quality assurance tools can be employed to maintain a high quality of care? We developed a machine learning algorithm to assess the presences of motion artifacts in low-dose CT scans of the chest.

A Machine Learning Model Based on the National Lung Cancer Screening Trial to Aid Image Quality Analysis: A Feasibility Study
A.D. Brown & D.P. Deva. Canadian Association of Radiologists the 80th Annual Scientific Meeting