# CT-based radiomics in predicting the efficacy of preoperative neoadjuvant chemoimmunotherapy for non-small cell lung cancer: a systematic review and meta-analysis

## TL;DR

This study reviews how CT-based radiomics can predict how well lung cancer patients will respond to pre-surgery chemoimmunotherapy treatment.

## Contribution

The study provides a meta-analysis of CT-based radiomics models for predicting treatment outcomes in non-small cell lung cancer patients.

## Key findings

- CT-based radiomics models have a pooled AUC of 0.81 for internal validation and 0.80 for external validation.
- Deep learning models outperformed machine learning in sensitivity and specificity for predicting treatment response.
- 2D ROI-based models showed higher sensitivity and specificity compared to other feature selection methods.

## Abstract

Neoadjuvant chemoimmunotherapy significantly improves surgical resection rates, major pathological response rates (MPR), pathological complete response rates (pCR), and survival rates in patients with resectable NSCLC. Through systematic reviews and meta-analyses, we examined the diagnostic value of CT-based predictive models in predicting neoadjuvant chemoimmunotherapy treatment outcomes for NSCLC.

PubMed, Embase, Web of Science databases, China National Knowledge Infrastructure, and Wanfang were systematically searched up to January 12, 2026. To assess study risk of bias and quality, we employed the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) tool and the Radiomics Quality Score version 2.0(RQS). Diagnostic accuracy of radiomics for detecting neoadjuvant chemoimmunotherapy pathological response in NSCLC patients was evaluated by calculating the area under the curve (AUC), sensitivity, specificity, and accuracy for each study.

The meta-analysis analyzed 17 studies with 4,510 individual subjects. The pooled AUC, sensitivity, and specificity of internal validation models were 0.81, 0.79, and 0.69, respectively. The pooled AUC, sensitivity, and specificity of external validation models were 0.80, 0.75, and 0.73, accordingly. Subgroup analyses revealed that models using deep learning (DL) algorithms demonstrated superior sensitivity (internal: 0.79, 95% CI: 0.73-0.85; external: 0.77, 95% CI: 0.72-0.82) and specificity (internal: 0.79, 95% CI: 0.74-0.85; external: 0.73, 95% CI: 0.68-0.78) compared to those using machine learning (ML). Models predicting MPR exhibited higher sensitivity in internal validation (0.82, 95% CI: 0.77-0.86), while showing higher specificity in external validation (0.76, 95% CI: 0.72-0.81). In contrast, models predicting pCR demonstrated the opposite pattern. Features selected using the intraclass correlation coefficient (ICC) demonstrated significantly higher pooled sensitivity (internal: 0.85, 95% CI: 0.80-0.89; external: 0.81, 95% CI: 0.76-0.87) and specificity (internal: 0.70, 95% CI: 0.63-0.78; external: 0.77, 95% CI: 0.71-0.82) compared to non-ICC-selected features. When stratified by the median Radiomics Quality Score (RQS ≥ 41.07%), higher-scoring studies were associated with lower pooled sensitivity (internal: 0.78, 95% CI: 0.73-0.84; external: 0.71, 95% CI: 0.66-0.76) but a trend toward higher specificity. Finally, models based on two-dimensional regions of interest (2D ROI) demonstrated higher pooled sensitivity (internal: 0.86, 95% CI: 0.80-0.92; external: 0.87, 95% CI: 0.79-0.96) and specificity in external validation (0.80, 95% CI: 0.68-0.91).

Due to its good diagnostic accuracy, widespread use, and low cost, CT-based radiomics can be used to predict the efficacy of neoadjuvant chemoimmunotherapy in NSCLC preoperatively.

https://www.crd.york.ac.uk/prospero/, identifier (CRD420251174128).