The pressures imposed by a rapidly expanding knowledge base, gaps in training, limited experience and escalating time constraints create a dilemma for diagnostic radiologists and their patients. How can radiologists consistently provide quality diagnostic interpretations? How can radiologists reduce intra- and interobserver variability in interpretation such that the diagnosis is less dependent on the interpreting radiologist and more accurately reflects the presence or absence of the underlying disease? Our long-term objectives are (1) to develop a general statistical methodology for the development and implementation of a decision supporting system (DSS) to help physicians to make informed decisions in radiologic diagnosis and to reduce intra- and interobserver variability and (2) to develop a new general statistical inferential framework that can determine if the performance of our DSS is equivalent to that of an expert or a panel of experts. Our immediate goal and proof of concept is motivated by the need to develop a DSS to improve the care of nephrology patients referred for a nuclear medicine renal scans, an area where many radiologists lack both training and experience. A renal scan is obtained by injecting a radioactive tracer, 99mTc MAG3 and sequentially imaging that tracer over a 20-30 min period as it is removed from the blood by the kidneys and passes down the ureters into the bladder. When obstruction is suspected, the patient often receives a potent diuretic and sequential images over the kidney are obtained for an additional 20 min. Radiologists typically use a few specific points on the kidney time activity curves (renogram) to assist in interpretation of the study. We propose to integrate clinical data with automated image analysis to provide a comprehensive interpretation of MAG3 renal scans in a structured format. Rather than using a few isolated features on the renogram, we propose to develop a latent class modeling approach for predicting kidney obstruction that jointly models renogram curve data (functional data [13,49]) resulting from renal images and expert ratings as well as other relevant clinical variables (Aim 1). Extensions will be developed for handling missing data that are present in this type of studies. In order to evaluate the newly developed DSS, we propose to develop a new general statistical inferential framework that can determine if the performance of our DSS is equivalent to that of an expert or a panel of experts. The methodology is developed for both categorical and continuous ratings of the disease status (Aim 2). We plan to validate the DSS with an independent data sample (Aim 3). [We plan to conduct two pilot studies with (a) nuclear medicine residents and (b) radiology residents to determine the feasibility of applying DSS to clinical setting under Aim 4]. While intended to be of direct benefit to the interpretation of renal scans, the DSS and statistical methodology to be developed address common and fundamental issues in image interpretation, especially where the integration of data is needed to recover the information about the disease.

Public Health Relevance

We propose to develop statistical methods to derive a Decision Supporting System in radiologic diagnosis, particularly in areas where time stressed radiologists have limited experience and training. We plan to conduct two pilot studies with residents to evaluate the feasibility of applying DSS in clinical setting. These methods will help to educate physicians to make informed decisions in radiologic diagnosis while reducing intra- and inter- observer variability and facilitating faster interpretation with higher level of accuracy.

Agency
National Institute of Health (NIH)
Institute
National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK)
Type
Research Project (R01)
Project #
5R01DK108070-05
Application #
9987593
Study Section
Biostatistical Methods and Research Design Study Section (BMRD)
Program Officer
Chan, Kevin E
Project Start
2016-09-15
Project End
2021-07-31
Budget Start
2020-08-01
Budget End
2021-07-31
Support Year
5
Fiscal Year
2020
Total Cost
Indirect Cost
Name
Emory University
Department
Biostatistics & Other Math Sci
Type
Schools of Public Health
DUNS #
066469933
City
Atlanta
State
GA
Country
United States
Zip Code
30322
Chang, Changgee; Kundu, Suprateek; Long, Qi (2018) Scalable Bayesian variable selection for structured high-dimensional data. Biometrics :
Taylor, Andrew T; Brandon, David C; de Palma, Diego et al. (2018) SNMMI Procedure Standard/EANM Practice Guideline for Diuretic Renal Scintigraphy in Adults With Suspected Upper Urinary Tract Obstruction 1.0. Semin Nucl Med 48:377-390
Taylor, Andrew T; Folks, Russell D; Rahman, A K M Fazlur et al. (2017) 99mTc-MAG3: Image Wisely. Radiology 284:200-209