(주) 뷰노

JRC 2025
-

Deep Learning-based Slice Thickness Reduction for Lung Nodule Detection of Thick Slice Chest CT

  • Apr. 2025

Abstract

Computer-aided detection (CAD) systems for lung nodule detection often take challenges with 5-mm computed tomography (CT) scans, leading to missed nodules. We explores the efficacy of a deep learning-based algorithm to improve slice thickness, aiming to enhance the lung nodule diagnostic performance of CAD systems. This retrospective study collected a total of 832 chest CT scans. The ground truths were defined based on a reader study by two radiologists and were finally confirmed by a third radiologist. In total, 704 nodules were annotated. We developed a deep-learning based SR algorithm, which was trained to convert 5mm thick slices into enhanced 1mm SR slices. The diagnostic performance of the CAD system was evaluated on both the original (5mm) and improved scans (1mm). The nodule-based analysis involved the Free-Response Receiver Operating Characteristic (FROC) curve, assessing sensitivity per 1 false positives (FPs) per scan. The case-based analysis utilized the Receiver Operating Characteristic (ROC) curve, calculating the Area Under the ROC Curve (AUROC). Sensitivity increased from 0.598 to 0.707 at 1 FPs/scan, and the AUROC improved from 0.761 to 0.814. The qualitative analysis further supported these results, showing enhanced visibility and characterizations of nodules. Visual evaluations revealed that nodules undetected in 5mm scans were consistently identified in the SR 1mm scans. Applying a deep learning-based algorithm for slice thickness improvement from 5mm to 1mm enhances the CAD system's performance in detecting lung nodules on low-dose chest CT scans.

Tag in

#medical_image

#VUNO Med®-LungCT AI™