Radiomics and Machine Learning in Medical Imaging / Advanced X-ray and Ct Imaging / MRI in Cancer Diagnosis · Journal article

Noninvasive prediction of classic EGFR-activating mutations in non-small cell lung cancer using a combined spectral computed tomography model: an auxiliary tool for preoperative stratification of candidates for EGFR-TKI therapy

BMC Pulmonary Medicine · August 11, 2026

Promising Why this matters to you Expert discussion

Encouraging direction, but not yet definitive.

AI Summary

A retrospectively developed spectral CT diagnostic model combining clinical and imaging features achieved an AUC of 0.805 for predicting EGFR-activating mutations in NSCLC, significantly outperforming conventional CT alone. The model showed internal consistency and low overfitting but lacks external validation and prospective confirmation needed to establish clinical utility.

What was studied

Retrospective cohort study with internal validation. NSCLC patients undergoing preoperative assessment; eligibility criteria and exclusion criteria not detailed in abstract.. Intervention: Dual-layer spectral CT imaging with clinical and conventional CT data integration into predictive model. Compared with: Clinical-conventional CT model (without spectral CT parameters). n = 129. Not stated in abstract.

Key findings

Combined spectral CT model achieved AUC of 0.805 for EGFR mutation prediction Combined model significantly outperformed clinical-conventional CT model (AUC 0.642, P < 0.001) Venous-phase normalized iodine concentration was the only independent predictor (OR = 2.578, P = 0.018)

Safety

Safety was not reported in the material analysed. Check the source before drawing any conclusion about harm.

Who this may help

This model may assist preoperative stratification for EGFR-TKI candidacy and reduce reliance on invasive biopsy; however, external validation in an independent prospective cohort is required before routine clinical adoption can be recommended.

Plain-language summary

A single-centre retrospective study with internal validation showing a diagnostic model for EGFR mutation prediction; the result is clear and clinically relevant but limited by retrospective design, modest sample size, and lack of external validation.

Study design

As stated by the source record.

MethodsRetrospectively enrolled 129 NSCLC patients and developed a combined model integrating clinical features, conventional CT findings, and dual-layer spectral CT parameters. Model performance evaluated using ROC analysis, DeLong test, decision curve analysis (DCA), and 1000-iteration bootstrap validation for overfitting assessment.

DesignRetrospective cohort study with internal validation

PopulationNSCLC patients undergoing preoperative assessment; eligibility criteria and exclusion criteria not detailed in abstract.

Sample size129

InterventionDual-layer spectral CT imaging with clinical and conventional CT data integration into predictive model

ComparatorClinical-conventional CT model (without spectral CT parameters)

SettingNot stated in abstract

Reported figures

Quoted from the source exactly as published.

AUC (combined model)0.805

AUC (clinical-conventional CT model)0.642

P-value (model comparison)< 0.001

Odds ratio (venous-phase NIC)2.578

P-value (venous-phase NIC)0.018

AUC (adenocarcinoma subgroup)0.810

Sample size129

Clinical bottom line

Caveats

Single-centre retrospective study lacks external validation in an independent cohort
No prospective validation or comparison with actual clinical decision-making outcomes reported
Sensitivity, specificity, and positive/negative predictive values not reported
No head-to-head comparison with other noninvasive EGFR prediction methods

Evidence assessment

Graded across the dimensions that decide whether you should act, each from what the source actually supports. There is no single score, and where a dimension was not assessed it says so.

Study design

moderate

Retrospective cohort with internal validation; no external validation or prospective confirmation.

Directness

strong

Directly predicts EGFR mutation status using noninvasive imaging for clinical decision-making.

Consistency

limited

Single-centre study; no comparison with independent cohort or prior models reported.

Risk of bias

moderate

Retrospective design and internal-only validation increase risk; 1000-iteration bootstrap mitigates overfitting concern.

Sample adequacy

limited

129 patients total; adequate for model development but modest for robust generalization claims.

Source abstract

Abstract Background EGFR genotyping is essential for EGFR-TKI therapy in non-small cell lung cancer (NSCLC), but invasive biopsy has limitations. Noninvasive preoperative stratification tools are urgently needed. Methods We retrospectively enrolled 129 NSCLC patients to develop and internally validate a combined model integrating clinical features, conventional CT findings, and dual-layer spectral CT parameters for predicting classic EGFR-activating mutations. Performance was evaluated by ROC, DeLong test, DCA, and 1000-iteration bootstrap validation. Results The combined model achieved an AUC of 0.805, significantly outperforming the clinical-conventional CT model (AUC = 0.642, P < 0.001). Venous-phase normalized iodine concentration (NIC) was the only independent predictor (OR = 2.578, P = 0.018). The model showed robust performance in the lung adenocarcinoma subgroup (AUC = 0.810) with negligible overfitting. Conclusions This noninvasive spectral CT model provides a reliable, easy-to-implement auxiliary tool for preoperative EGFR mutation stratification to optimize EGFR-TKI therapy decisions.

Follow from this

Taken from the source record, never inferred. Follow any of these and new work involving them reaches your briefing.

Radiomics and Machine Learning in Medical ImagingXueYan LiaoHui LinQianqian ZhanAdvanced X-ray and Ct ImagingMRI in Cancer DiagnosisDandan Lin