Document is current

Article History

Received: 29 November 2024

Accepted: 4 May 2026

First Online: 12 May 2026

Declarations

:

: This study was conducted in strict adherence to the principles of the Declaration of Helsinki. The study protocol was reviewed and formally approved by the Ethics Committee of the Fifth Affiliated Hospital of Sun Yat-sen University (approval number: K107-1). Given the anonymous and retrospective nature of this study, the Ethics Committee waived the requirement for written informed consent from the participants. All data were processed anonymously to ensure the confidentiality and privacy of the participants.

: Not applicable.

: Not applicable.

: This study adheres to the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) statement. The checklist items and their compliance are as follows: Title and Abstract: The study is clearly stated as a multi-modal deep learning model for three-class classification (AIS, MIA, IAC) of GGNs in lung adenocarcinoma. The abstract includes background, methods, results, and conclusions, meeting the requirements. Background and Objectives: The clinical need for differentiating GGN subtypes and the limitations of existing studies are clearly elaborated. The study objective (developing and validating a multi-modal fusion model) is clear, meeting the requirements. Study Design: A retrospective multi-center study involving 4 medical centers is clearly defined. The inclusion/exclusion criteria and sample size determination basis are detailed, meeting the requirements. Study Population: A total of 431 GGN samples (60 AIS, 57 MIA, 314 IAC) are included, with clear sources, baseline characteristics, and grouping methods (training/validation/test sets), meeting the requirements. Predictor Variables: CT image features (extracted by ResNet50), 8 clinical variables, and serum tumor markers are clearly included. The measurement methods and preprocessing of variables are detailed, meeting the requirements. Outcome Variables: The outcome is defined as three-class classification of AIS, MIA, and IAC, with pathological confirmation as the gold standard. The definition is clear, meeting the requirements. Data Preprocessing: CT image cropping, normalization, encoding and standardization of clinical variables, and data augmentation strategies (only applied to the training set) are detailed, meeting the requirements. Model Development: The ResNet50 feature extraction process, multi-modal feature fusion method (vector concatenation), classifier (RBF-SVM), and hyperparameter optimization methods are clearly defined, meeting the requirements. Model Validation: 5-fold cross-validation is adopted, with a clearly set independent test set. Various evaluation indicators (ACC, AUC, Precision, Recall, etc.) are detailed, meeting the requirements. Model Performance: The performance of the model is clearly compared with radiologists of different experience levels, with per-class and overall indicators provided and reasonably interpreted, meeting the requirements. Model Interpretability: Heatmap/segmentation visualization is not included currently (limitations are explained), and Grad-CAM will be added in future studies to improve interpretability, meeting the requirements. Limitations: Key limitations (lack of external validation, no radiomic features included) are clearly pointed out, with targeted improvement directions proposed, meeting the requirements. Conclusions: The clinical value of the model is clearly stated without overstating performance, which is consistent with the actual study, meeting the requirements. Data Availability: Data are not publicly available due to privacy restrictions but can be obtained from the corresponding author upon reasonable request, meeting the requirements.

: The authors declare no competing interests.