(주) 뷰노

JRC 2025

Deep Learning Based Computer Aided Diagnosis System for Usual Interstitial Pneumonia in Chest HRCT

  • Apr. 2025

Abstract

Usual Interstitial Pneumonia (UIP) is a lung disease that is difficult to ac
curately diagnose with chest computed tomography (CT) scans due to its compl
ex pattern. This study aims to develop a Computer Aided Diagnosis (CAD) Syst
em utilizing Convolutional Neural Networks (CNNs) to enhance the precision o
f UIP identification and decision-making. We collected a substantial dataset
 of chest CT scans from 1205 patients with UIP and non-UIP cases at Korean t
ertiary Hospital. We obtained 503 cases with axial HRCT (381 UIP cases, and
122 non-UIP cases). We utilized the CNN-based model for classification to UI
P and non-UIP cases, and compared its results with the honeycomb segmentatio
n masks from our pre-existing Lung Lesion Quantification model. Our model wa
s assessed using various performance metrics, including AUROC, sensitivity,
specificity, and accuracy. We achieved an AUROC of 0.77 for distinguishing b
etween UIP and non-UIP cases. The sensitivity and specificity were 0.800 and
 0.575, and an accuracy of 0.688. In this study, we developed an deep learni
ng based CAD for UIP classification from chest CT scans using CNNs. The robu
stness and high accuracy offer promising prospects for clinical practice. It
 has the potential to assist radiologists and clinicians in making more accu
rate and timely diagnoses of UIP, ultimately improving patient outcomes. To
fully utilize this system's potential in improving UIP diagnosis, further va
lidation and integration of this system into routine clinical workflows is required.

Link

-

Author

Jonghun Jeong, Jung-Hyun Kang, Jeongmin Kim, Kyoung-Min Moon

Tag in

#medical_image

#VUNO Med®-LungCT AI™