Objective:

To develop and validate a foundation model–driven multimodal fusion framework for non-invasive prediction of treatment response to first-line chemo-immunotherapy in patients with advanced lung squamous cell carcinoma (LUSC).

Approach:

• Data Collection: Baseline contrast-enhanced CT images and clinical data were collected from patients with advanced LUSC receiving first-line chemo-immunotherapy.
• Feature Extraction: Handcrafted radiomics features were extracted from tumor regions, and 2.5D deep learning features were extracted using a DINO-pretrained vision Transformer foundation model.
• Model Construction: A multi-source features fusion model was constructed using machine learning classifiers, incorporating both radiomics and clinical variables.
• Model Evaluation: Model performance was evaluated using AUC, accuracy, sensitivity, and specificity, along with DeLong testing, DCA, NRI, and IDI for comparative assessment.
• Interpretability Analysis: SHAP analysis was applied to enhance model interpretability and decision transparency.

Key Findings:

• The multi-source features fusion model (FusionModel) achieved an AUC of 0.903 and an accuracy of 0.885 in the training cohort.
• In the validation cohort, FusionModel achieved an AUC of 0.863 and an accuracy of 0.836.
• FusionModel outperformed single-modality feature models and demonstrated superior net clinical benefit on DCA.
• NRI and IDI analyses indicated improved reclassification ability.
• SHAP analysis showed that deep learning features were the dominant contributors, with radiomics and clinical variables providing complementary information.

Interpretation:

The foundation model-driven multi-source features fusion model enabled accurate prediction of chemo-immunotherapy response in advanced LUSC.

Limitations:

• The study is retrospective and may be subject to selection bias.
• The generalizability of the findings may be limited to the specific patient cohort studied.

Conclusion:

The study presents a non-invasive tool for treatment stratification in advanced LUSC.