This is the authors' abstract. We don't add key points for this paper.
Objective Lung cancer is one of the main causes of cancer-related deaths worldwide. To improve early diagnosis and treatment, it is essential to accurately classify its subtypes. This study introduces a new method that combines radiomic and deep learning features to classify lung cancer subtypes, addressing challenges like tumor diversity and differences in imaging quality. Materials and methods The study used data from 2569 patients covering five lung cancer subtypes: Adenocarcinoma, Squamous Cell Carcinoma, Large Cell Carcinoma, Small Cell Lung Cancer, and Pulmonary Carcinoid Tumors. CT scans were processed to standardize voxel sizes and normalize image intensity. Tumor regions were identified and resized to a uniform size of 128 × 128 × 64.Radiomic features, such as intensity, texture, and shape, were extracted following international standards. Deep learning features were obtained using a 3D convolutional autoencoder with attention mechanisms, which helped focus on the most important tumor areas. Three feature selection methods—RFE, Boruta, and correlation analysis—were used to narrow down the features to 10–15 per method. Classification models, including CatBoost, XGBoost, SVM, and a ResNet-based CNN, were trained and tested with these features. A hybrid model combining radiomic and deep features was also evaluated. Results The hybrid model performed best, achieving 94 % accuracy and an AUC (Area Under the Curve) of 0.97 using CatBoost. This was better than using only radiomic features (87 % accuracy) or only deep learning features (89 % accuracy). The ResNet-based CNN achieved 91 % accuracy and an AUC of 0.95 when trained on the tumor regions directly. Combining radiomic and deep features improved the models' reliability by reducing overfitting and making them more generalizable. Feature selection was key in improving performance while keeping the models computationally efficient. Conclusions This study shows that combining radiomic and deep learning features can accurately classify lung cancer subtypes. The framework is scalable and adaptable, making it promising for use in diagnosing other types of cancer and in different imaging techniques. This advancement can help improve precision medicine in oncology.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Leng et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: