2024 Asia Conference on Lung Cancer (ACLC) - Poster & ePosters-EP02.03

EP02.03 - Feng Ying

Pdf Summary

The study conducted by Ying Feng from Chongqing Medical University investigates the diagnostic value of LungPro virtual navigation bronchoscopy (VBN) combined with pathological histology analysis for detecting peripheral lung cancer. Although traditional VBN is minimally invasive, its diagnostic efficiency for peripheral lung cancer remains limited. This study seeks to enhance diagnostic accuracy by integrating VBN with machine learning algorithms that assess pathological histology.<br /><br />Between January 2022 and December 2023, 142 patients underwent VBN and biopsy procedures, providing clinical and imaging data, alongside hematoxylin-eosin (HE) stained whole slide images. The factors influencing lung cancer diagnosis via VBN were assessed using univariable and multivariable logistic regression. The study utilized a multi-instance learning approach to derive WSI features and employed Lasso regression for selecting significant pathological characteristics. By incorporating these features into a machine learning diagnostic model, the researchers were able to differentiate between benign and malignant cases.<br /><br />Key findings indicate that VBN is particularly effective in scenarios involving solid lesions, CT-confirmed bronchial signs, and lesions involving three or more bronchi. The integration of this technique with machine learning-enhanced pathological histology boosts the VBN diagnostic accuracy. The analysis revealed that such integration could achieve a diagnostic efficiency of up to 90.3% for peripheral lung cancer. Notably, in a cohort of 28 lung cancer patients initially yielding negative biopsy results, the model identified malignant characteristics in 20 cases.<br /><br />The study advocates for the expansion of datasets to further evaluate the applicability and clinical significance of these findings across a broader range of VBN samples. By doing so, it aims to validate and possibly increase the method's utility in clinical settings.

Keywords

LungPro virtual navigation bronchoscopy

peripheral lung cancer

pathological histology analysis

machine learning algorithms

multi-instance learning

Lasso regression

diagnostic accuracy

hematoxylin-eosin stained images

bronchial signs

clinical significance