Detection of Lung Parenchymal Lesions in a Low-Prevalence Clinical Setting Using Deep Learning

KVAK, Daniel, Martin ČERNÝ and Jakub DANDÁR. Detection of Lung Parenchymal Lesions in a Low-Prevalence Clinical Setting Using Deep Learning. In European Congress of Radiology 2024, Vienna. 2024.

Basic information

• Original name: Detection of Lung Parenchymal Lesions in a Low-Prevalence Clinical Setting Using Deep Learning
• Authors: KVAK, Daniel, Martin ČERNÝ and Jakub DANDÁR.
• Edition: European Congress of Radiology 2024, Vienna, 2024.

Other information

• Type of outcome: Conference abstract
• Confidentiality degree: is not subject to a state or trade secret
• Keywords in English: Artificial Intelligence, Computer applications, Lung, CAD, Digital radiography, PACS, CAD, Computer Applications-Detection, diagnosis, Decision analysis, Cancer
• Tags: International impact, Reviewed

Abstract

In the area of diagnostic imaging, the chest X-ray (CXR) is a fundamental modality routinely used for diagnostic assessments, particularly in the identification of lung abnormalities. However, the reliability of CXR is often questioned due to limitations of spatial resolution and the diverse nature of human anatomy, with the difficulty of distinguishing diagnostic errors from opinion discrepancies among observers. The aim of this study is to evaluate the performance of an automatic deep learning-based detection (DLAD) algorithm in detecting pulmonary lesions on CXR images, particularly in a clinical setting characterized by low disease prevalence. As computer-aided detection (CAD) systems increasingly move from experimental research into clinical practice, this study serves as a critical examination of their effectiveness, specifically in real-world scenarios that differ from balanced conditions in a simulated environment, i.e., equal numbers of normal and abnormal scans [1,2].