Modelo de deep learning para análise de nódulos pulmonares em imagens médicas

Abstract

Lung cancer is one of the leading causes of cancer-related deaths worldwide, and early diagnosis is crucial for improving patient survival rates. Chest computed tomography scans are essential tools for detecting pulmonary nodules, which may be indicative of lung cancer. However, manual analysis of these images by radiologists is time-consuming and subject to inter-observer variability. In this perspective, this study presents software for the analysis of medical imaging exams using machine learning. These images are derived from chest tomography scans. The developed software performs the segmentation process of potential pulmonary nodules present in the image and classifies them as nodules or non-nodules, serving as a diagnostic aid tool. The development was carried out in Python language, using the PyTorch library to implement the convolutional neural network. The LIDC-IDRI database was used, which contains over 800 chest CT scans with annotations made by radiologists about the nodules present. As a result, the system demonstrated the ability to segment and classify 85% of the pulmonary nodules present in exams from the dataset accurately.