Integrating quantitative and convolutional features to enhance the efficiency of pathology classification in CT imaging

Intellectual Systems and Technologies

The paper proposes an approach that combines radiomic features and deep learning to enhance the accuracy of image classification obtained from lung computed tomography (CT) scans. The deep convolutional neural network ResNet18 was used to extract convolutional features from CT images. Radiomic features describing texture, shape, and intensity were combined with these convolutional features to improve the feature description of the lung CT image dataset. Using Principal Component Analysis (PCA) and feature selection methods, the most informative set of 250 features was obtained. Machine learning models, including Random Forest and Support Vector Machines (SVM), were used for classification. The SVM classifier showed the best results, achieving a classification accuracy of 0.97. The addition of genetic data allowed an improvement in classification accuracy. The study underscores the importance of combining advanced computational methods and data processing methodologies to solve image classification tasks.