Metodología machine learning para el tratamiento de imágenes computarizadas en pacientes con cancer de pulmon

Abstract

This research proposes a computational methodology aimed at identifying patterns associated with lung cancer, using exclusively machine learning and deep learning tools implemented in Python. The study is based on the analysis of the LIDC-IDRI dataset (The Lung Image Database Consortium and Image Database Resource Initiative), provided by the U.S. National Cancer Institute, which contains medical images in DICOM (Digital Imaging and Communications in Medicine) format. DICOM is the international standard for the transmission, storage, and processing of medical images, allowing the integration of patient information, acquisition characteristics, and the image itself into a single file. In addition to DICOM images, the dataset includes radiologist segmentations, nodule counts, and clinical diagnoses in structured files. This methodology focuses on the processing, integration, and analysis of large volumes of data, with the aim of exploring significant correlations and behaviors within the available variables. Although the purpose is not to provide direct clinical diagnosis, the patterns identified could serve as a basis for future research and support the development of diagnostic assistance systems. Each patient can generate between 10 and 15 GB of information, which poses relevant challenges regarding efficient processing, organization, and data interpretation. This work seeks to contribute to strengthening computational analysis applied to lung cancer, from an engineering, exploratory perspective, centered on leveraging complex medical data.