# Integration of dual-energy CT parameters and radiomics features for non-invasive prediction of α-SMA and CD8 + T cell in non-small cell lung cancer

## TL;DR

This study combines dual-energy CT and radiomics to non-invasively predict fibrosis and immune cell infiltration in lung cancer, potentially guiding immunotherapy decisions.

## Contribution

A novel integration of dual-energy CT parameters and radiomics features for non-invasive assessment of α-SMA and CD8+ T cells in NSCLC.

## Key findings

- An integrated model combining DECT and radiomics features achieved an AUC of 0.766 for predicting high α-SMA expression.
- Normalized iodine concentration and spectral slope of K40-70 were key features associated with α-SMA and CD8+ T-cell infiltration.
- DECT-only models performed comparably to radiomics models for predicting CD8+ T-cell density without significant improvement from integration.

## Abstract

The non-invasive characterization of the tumor microenvironment (TME) is essential for stratifying non-small cell lung cancer (NSCLC) patients who may benefit from immunotherapy. This study investigates a novel approach by integrating dual-energy CT (DECT) parameters with radiomics to quantitatively assess stromal fibrosis (via α-SMA area) and CD8 + T-cell infiltration.

In this prospective study, 70 treatment-naive NSCLC patients were enrolled. Preoperative DECT scans were used to extract both DECT parameters and radiomics features. Corresponding surgical specimens were analyzed to determine the area percentage of α-SMA-positive stroma and the density of CD8 + T cells, with patients classified into high and low groups for each biomarker. After feature selection, models were constructed based on DECT parameters alone, radiomics features alone, and a combined feature set. Models were evaluated via 5-fold cross-validation.

For predicting high α-SMA expression, the integrated model combining DECT parameters and radiomics features demonstrated superior performance (AUC: 0.766) compared to models using either modality alone (DECT AUC: 0.670; radiomics AUC: 0.703). In contrast, for predicting CD8 + T-cell density, the DECT-only model (AUC: 0.715) performed comparably to the radiomics model (AUC: 0.695), with no significant gain from integration. Key discriminating features, such as normalized iodine concentration for α-SMA and spectral slope of K40-70 for CD8+, showed significant intergroup differences and plausible biological correlations.

The integration of DECT and radiomics presents a feasible, non-invasive strategy to assess specific TME components in NSCLC, underscoring the complementary value of different imaging data types towards developing biomarkers for personalized oncology.