Algoritmo para reducir características radiómicas en la clasificación de nódulos pulmonares
Year: 2026
Type: article
Abstract: Lung cancer is one of the leading causes of death. Computed tomography is the most widely used technique to diagnose and evaluate responses to treatment. Computer-aided diagnosis systems have proven effective in assisting radiologists in their decision making. Radiomics aims to automatically extract features from digital images and develop models to predict lesion phenotypes non-invasively. However, managing the redundancy and instability of these features remains a significant challenge, compromising model reliability and generalizability. In this study, a system based on reduced Radiomics-Machine Learning was established. An approach based on the calculation of the Pearson correlation coefficient between each radiomic feature and the database annotation was proposed to achieve sensitive and efficient feature reduction, avoiding redundancies and reducing computational cost compared to other algorithms in the state of the art. A set of 275 nodules from the annotated LIDC-IDRI database was used. A group of 102 radiomic features was extracted and reduced to just 5 for binary classification and 3 for classification into five degrees of malignancy. Ten models of Support Vector Machine and Random Forest were trained and validated. In both experiments, the best models achieved sensitivity, specificity, precision, and accuracy values above 90%, surpassing the values achieved by different approaches in scientific literature for the same database. It is concluded that the proposed feature reduction methodology contributes to a more robust and reliable classifier with exceptionally low computational cos.t
Language: Spanish
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7143262939