Skip to Main Content

An official website of the United States government

Identification of actionable lung nodules with low-dose computed tomography screening: Characterizing the effects on observer performance of a personalized Artificial Intelligence feedback

Principal Investigator
Name
Jessica Sieren
Degrees
PhD
Institution
University of Iowa
Position Title
Associate Professor
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-895
Initial CDAS Request Approval
Mar 21, 2022
Title
Identification of actionable lung nodules with low-dose computed tomography screening: Characterizing the effects on observer performance of a personalized Artificial Intelligence feedback
Summary
As a result of the expanded recommendation regarding who to screen for lung cancer, the number of Low Dose Computed Tomography (LDCT) cases is supposed to increase by 85% in the US, from approximately 227,000 to approximately 419,000 examinations over the lifetime of 100,000 screened individuals. However, on average radiologists’ sensitivity for detecting actionable lung nodules in LDCT scans is only 49%, as their visual search strategies only cover on average 27% of the lungs’ parenchyma. In this project we will use eye position recording and an Artificial Intelligence algorithm to offer personalized feedback to radiologists about the presence of lung cancer in the areas of the parenchyma not scanned by them.
Aims

As a result of the expanded recommendation regarding who to screen for lung cancer, the number of Low Dose
Computed Tomography (LDCT) cases is anticipated to increase by 85% in the US, from approximately 227,000 to
approximately 419,000 examinations over the lifetime of 100,000 screened individuals. However, on average
radiologists' sensitivity for detection actionable lung nodules in LDCT scans is only 49%, as their visual search
strategies only cover on average 27% of the lungs' parenchyma. In this project, we will use eye position recording
and an Artificial Intelligence algorithm to offer personalized feedback to radiologists about the presence of lung
nodules in the areas of the parenchyma not scanned by them.
SA1: Understand and quantify radiologists' reading search pattern of LDCT images for the detection of lung nodules
without AI assistance.
SA2: Understand and quantify the effects of personalized AI on the performance of radiologists reading LDCT for
detection of lung nodules.

Collaborators

Dr Claudia Mello-Thoms, University of Iowa