2024 Asia Conference on Lung Cancer (ACLC) - Poster & ePosters-EP02.02

EP02.02 - Tingyue Luo

Pdf Summary

The study aimed to construct a differential diagnosis model using CT-based radiomics to differentiate immune-related interstitial pneumonia (CIP) from other types of pneumonia. CIP is a severe adverse reaction occurring after treatment with immune checkpoint inhibitors, posing significant risks if not diagnosed and treated promptly. Currently, CIP diagnosis relies on clinician expertise and invasive procedures like lung biopsies, leading to frequent misdiagnoses.<br /><br />This research developed a model using machine learning to analyze lung CT images, enhancing the accuracy and efficiency of non-invasive CIP diagnosis. The experimental group comprised 72 patients with CIP-related CT images, while the control group had 69 patients without CIP after immune checkpoint inhibitor use. From these images, 258 significant imaging features were identified, which were used to construct a Support Vector Machine (SVM) model. The performance of the SVM model was measured with an area under the ROC curve (AUC) of 0.77.<br /><br />To improve diagnostic accuracy, a regression model integrating clinical characteristics was constructed. This enhanced model outperformed the SVM model in the validation dataset, achieving a higher diagnostic rate with an AUC of 0.98. The study concluded that the non-invasive regression model significantly improves the diagnosis of CIP, showcasing the potential of artificial intelligence in the differential diagnosis of diseases. This advancement may ultimately lead to better patient outcomes and clinical decision-making without the need for invasive diagnostics.

Keywords

CT-based radiomics

immune-related interstitial pneumonia

CIP diagnosis

immune checkpoint inhibitors

machine learning

lung CT images

Support Vector Machine

ROC curve

non-invasive diagnosis

artificial intelligence