Development and validation of an interpretable prediction model using spatial patterns of tumor-infiltrating lymphocytes in H&E-stained whole-slide images for immune subtyping of lung adenocarcinoma - Report - MDSpire
Advertisement
Development and validation of an interpretable prediction model using spatial patterns of tumor-infiltrating lymphocytes in H&E-stained whole-slide images for immune subtyping of lung adenocarcinoma
Clinical Report: Predictive Model for Immune Classification of Lung Adenocarcinoma
Overview
This study presents a transparent predictive model for immune subtyping of lung adenocarcinoma (LUAD) based on the spatial distribution of tumor-infiltrating lymphocytes (TILs) in H&E whole-slide images. The model demonstrates high predictive performance, achieving AUCs of 0.839 and 0.927 in internal and external validation cohorts, respectively.
Background
Lung adenocarcinoma (LUAD) is the most prevalent subtype of non-small cell lung cancer and is associated with poor prognosis, particularly in advanced stages. The immune microenvironment plays a crucial role in determining the efficacy of immunotherapy, which has a low overall response rate in LUAD patients. Understanding the spatial distribution of TILs can provide insights into immune status and guide treatment decisions.
Data Highlights
Metric
Internal Cohort
External Cohort
AUC for Immune Subtype Classification
0.839
0.927
Dice Score for Tissue Contour Segmentation
95.09%
N/A
F1-Score for TIL Identification
79.51%
N/A
Key Findings
503 LUAD patients were stratified into high- and low-immunity subgroups based on transcriptomic analysis.
The high-immunity group showed increased infiltration of CD8+ T cells and M1 macrophages.
Automated annotation model achieved a 95.09% Dice score for tissue contour segmentation.
The immune subtype prediction model utilized a 0.2 high-attention threshold for TIL spatial distribution analysis.
Immunohistochemical analysis confirmed higher densities of immune cells in high-immunity samples.
Clinical Implications
The developed model provides a cost-effective and scalable tool for assessing tumor immune status in LUAD, which can aid in identifying patients who may benefit from immunotherapy. The integration of deep learning with traditional histopathological analysis enhances the interpretability and applicability of immune classification in clinical settings.
Conclusion
This study establishes a novel and interpretable model for immune subtyping in LUAD, linking morphological features to immune profiles. The findings underscore the importance of TIL spatial distribution in evaluating the tumor immune microenvironment.
