Radiomics and Machine Learning in Medical Imaging / Lung Cancer Diagnosis and Treatment · Journal article

Predicting acquired T790M mutations in non-small cell lung cancer after EGFR-TKI therapy: integration of delta-radiomics and clinical characteristics with SHAP interpretation

BMC Medical Imaging · September 7, 2026

Preliminary Why this matters to you Expert discussion

Early or partial results. Treat as a signal, not a conclusion.

AI Summary

A retrospective, single-centre study developed a predictive model combining delta-radiomics and clinical features to forecast EGFR T790M resistance mutations in NSCLC patients treated with EGFR-TKIs. The combined model achieved test-set AUC of 0.82 and accuracy of 79.2%, but requires prospective multicentre validation before clinical deployment.

What was studied

Retrospective cohort study with machine learning model development and train-test validation. NSCLC patients with disease progression after first-line EGFR-TKI treatment; retrospective, single-centre cohort; January 2013 to September 2019.. Intervention: Delta-radiomics and clinical feature extraction; combined predictive model.. Compared with: Clinical model alone, NECT radiomics model alone, and delta-radiomics model alone.. n = 285. Single centre; geography not specified in source text..

Key findings

Combined delta-radiomics and clinical model achieved AUC 0.82 in test set (n=72) Combined model accuracy was 79.2%, compared to clinical model alone 66.7%, NECT radiomics 57.0%, and delta-radiomics alone 75.0% SHAP identified 10 significant features: 2 clinical and 8 radiomic features driving model decisions

Safety

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

Who this may help

A clinician might use this model to stratify risk of T790M-driven resistance prior to therapy, but the retrospective single-centre design and lack of prospective validation limit confidence in generalizability. The authors explicitly recommend multicentre prospective validation before clinical application.

Plain-language summary

Retrospective, single-centre study with modest sample size and surrogate endpoint (mutation status prediction); requires prospective multicentre validation before clinical use, as stated by the authors.

Study design

As stated by the source record.

MethodsPatients with progressed NSCLC after first-line EGFR-TKI treatment were retrospectively enrolled between January 2013 and September 2019. Patients were randomly assigned to training (n=213) and test (n=72) cohorts. Four models were developed using: clinical characteristics alone, non-contrast-enhanced CT radiomics alone, delta-radiomics features alone, and combined clinical and delta-radiomics features. Model performance was evaluated using AUC and decision curve analysis in the test set. SHAP interpretability was applied to identify and visualize key predictive features.

DesignRetrospective cohort study with machine learning model development and train-test validation

PopulationNSCLC patients with disease progression after first-line EGFR-TKI treatment; retrospective, single-centre cohort; January 2013 to September 2019.

Sample size285

InterventionDelta-radiomics and clinical feature extraction; combined predictive model.

ComparatorClinical model alone, NECT radiomics model alone, and delta-radiomics model alone.

SettingSingle centre; geography not specified in source text.

Reported figures

Quoted from the source exactly as published.

Test set AUC – combined model0.82

Test set AUC – clinical model0.70

Test set AUC – NECT radiomics model0.63

Test set AUC – delta-radiomics model0.75

Test set accuracy – combined model79.2%

Test set accuracy – clinical model66.7%

Test set accuracy – NECT radiomics model57.0%

Test set accuracy – delta-radiomics model72.2%

Clinical bottom line

Caveats

Single-centre study; generalizability to other populations or healthcare settings unknown.
Confidence intervals and p-values for AUC or accuracy not reported; statistical significance of model differences not tested.
Surrogate endpoint (mutation status prediction) used; impact on patient outcomes or treatment decisions not demonstrated.
Prospective validation required before clinical use, as stated by authors; no external validation cohort analysed.

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 random train-test split; no blinding or prospective follow-up reported.

Directness

moderate

Predicts T790M mutation status, a known resistance mechanism; clinical utility for treatment decisions not directly demonstrated.

Consistency

limited

Single-centre study; no comparison with external validation cohort or existing prediction tools reported.

Risk of bias

limited

Retrospective design; selection and measurement bias possible; data quality and loss-to-follow-up not described.

Sample adequacy

limited

Test set n=72; modest for multiclass radiomics model validation; confidence intervals for AUC not reported.

Source abstract

The aim of this study was to evaluate the value of combining delta-radiomics and clinical characteristics in predicting the risk of EGFR T790M resistance mutation in patients with non-small cell lung cancer (NSCLC) prior to first-line treatment with epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (EGFR-TKIs). This retrospective study included non-small cell lung cancer (NSCLC) patients whose disease progressed after first-line treatment with an EGFR tyrosine kinase inhibitor (TKI) between January 2013 and September 2019. The patients were randomly assigned to the training and test cohorts. Four predictive models for the acquisition of the T790M mutation were developed on the basis of clinical characteristics, non-contrast-enhanced computed tomography (NECT) radiomics features, delta-radiomics features, and a combination of clinical and delta-radiomics features. The optimal model was selected using the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA). Additionally, the Shapley additive explanations (SHAP) interpretability framework was applied to visualize and interpret the model’s decision-making process. A total of 285 patients were included in this study (213 in the training set and 72 in the test set). In the test set, the AUCs for the clinical model, NECT radiomics model, delta-radiomics model, and combined model were 0.70, 0.63, 0.75, and 0.82, respectively. The corresponding accuracies were 66.7%, 57.0%, 72.2%, and 79.2%, respectively. The SHAP tool identified 10 significant features, including 2 clinical features and 8 radiomic features. A delta-radiomics-based model combined with clinical features showed potential for predicting acquired T790M mutation status and provided preliminary interpretability, further prospective multicentre validation is needed before clinical application.

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 ImagingLei ShiChengting LinLung Cancer Diagnosis and TreatmentMingyue ZouXianglan ZhouLiTing Shi