2023 North America Conference on Lung Cancer (NACLC) - Posters and Abstracts-PP01.026 David Xiao Abstract

PP01.026 David Xiao Abstract

Pdf Summary

Radiomics, a field that involves analyzing radiographic images to extract quantitative features, has shown promise in improving the diagnosis of indeterminate pulmonary nodules (IPNs) in lung cancer. There are many platforms available for radiomic analysis, but there have been no comparisons between proprietary and open-source platforms. In this study, the researchers compared the performance of HealthMyne, a proprietary radiomic feature extractor, with PyRadiomics, an open-source feature extractor, in diagnosing lung cancer in IPNs.<br /><br />The researchers used HealthMyne to create a previously validated radiomic signature, and then used PyRadiomics to develop three different radiomic signatures. These signatures were based on different methods of feature selection. The models were trained on an internal cohort and externally validated on three other cohorts. The performance of the radiomic signatures was evaluated using the area under the receiver operating characteristic curve (AUC).<br /><br />The results showed that the proprietary radiomic signature had an AUC of 0.76, while the open-source signatures had AUCs of 0.71, 0.74, and 0.75, depending on the feature selection method. The researchers then combined the Mayo model score with the best-performing open-source signature and the proprietary signature to create clinical radiomic models (ClinRad). Both proprietary and open-source ClinRad models showed improved diagnostic accuracy compared to the Mayo model alone.<br /><br />The researchers concluded that the open-source platform had nearly identical diagnostic accuracy and improvement in risk stratification for IPNs compared to the proprietary platform. This suggests that the open-source platform is an accurate and accessible option for radiomic analysis. These findings have implications for the development and implementation of radiomics in lung cancer diagnosis.

Keywords

Radiomics

Indeterminate Pulmonary Nodules

Lung Cancer

HealthMyne

PyRadiomics

Radiographic Images

Feature Extraction

Feature Selection

Diagnostic Accuracy

Open-source Platform