Detecção automática de nódulos pulmonares utilizando transformers

Abstract

Lung cancer is the second most prevalent type of cancer in the world and the most deadly, accounting for one in five cancer deaths worldwide. The chances of survival of a patient detected with lung cancer increase considerably when the diagnosis is made early, with the 5-year survival rate reaching up to 70%. The lung cancer diagnosis is performed by radiologists through Computed Tomography (CT) images, but this is a complex task and subject to errors, since the nodules may be very small or located close to other anatomical structures. Professionals in the area also suffer from fatigue and a rush to diagnose, given that the number of exams to be analyzed is usually very high. Computer aided diagnosis (CAD) systems can automate this diagnosis process, reducing the time and effort from the specialists, as well as improving diagnostic confidence. Currently, deep learning (DL) techniques, especially with the use of CNNs, are the state-of-the-art for the automatic detection of pulmonary nodules, but with the introduction of the transformer architecture to computer vision tasks, a new area can be explored. In this context, the objective of this work is to present a system for the automatic detection of pulmonary nodules in CT images using the transformer architecture. We also evaluated the hypothesis that the transformer architecture is as efficient as the CNN models in the detection of pulmonary nodules.