Article Preview
Top1. Introduction
Radiodiagnosis disease detection is a low-cost and universally widespread method. Its main drawback is that it must be carried out by highly qualified people (a radiologist) which are scarce in the public health system. Thus, during high workload in a hospital, most X-ray images go directly to the doctor without being reviewed by a radiologist. For this reason, it is necessary to create automatic screening tools able to redirect suspicious cases to radiologists so that some diseases can be detected at earlier stages. In this paper, we will focus on screening lung nodules associated to neoformative processes of lung cancer.
At present, the simple chest radiograph continues to be the most performed radiological examination on a daily basis, being interpreted by numerous specialists. Among them, radiologists are trained specifically for the interpretation of such images, but they represent a small percentage of all the specialties between which medicine is divided, so many of these studies are not reported by a radiologist.
The vast majority of pathologies express noticeable radiological findings, easily detectable by any doctor, and who can consult the image with a radiologist. But the initial stage of the neoformative process of lung cancer, before becoming a lung mass, is represented on the chest X-ray as a pulmonary nodule, in most cases in a very subtle way. Even trained eyes can miss these findings, and it is not uncommon that once the pathology is diagnosed, by reviewing previous examinations, we can appreciate or imagine clues. Lung tumors present different cell lines, and present in different ways in the lung parenchyma. They may have a central arrangement, being located in perihilar or even endobronchial topographies, as well as a more peripheral distribution. Initially, they will be represented as small pulmonary nodules, always smaller than 3cm, and once that figure is exceeded, they will become pulmonary masses. Densotomographically, they can present densities similar to those of soft tissue tissues, presenting an enhancement after the administration of intravenous contrast (due to their greater vascularization and neoangiogenesis). But this representation can vary, since they can present a necrotic center, hemorrhagic foci or even intratumoral cystic components. They generally have poorly defined borders, with a poor definition with respect to the adjacent structures, adopting a morphology of “spiculate borders”, which as the disease progresses, can condition the invasion of organs and structures adjacent to the tumor. Given its variability in morphologies and locations where the pulmonary nodule can settle, it is important to recognize its presentation patterns both to identify them and to differentiate them from other pathologies (metastases, abscesses, interstitial diseases ... and much more).
For this reason, there are currently no programs for the early detection of lung tumors, as there are for other neoplasms (breast, prostate, colorectal, etc.), since computed tomography (CT) scans despite having greater sensitivity and specificity than Simple X-ray for the detection of the pulmonary nodule, administers high doses of radiation, making the risk / benefit in a large population unfavorable. And, given the aforementioned reasons to understand the difficulty of interpreting studies for the characterization of the pulmonary nodule, the simple chest X-ray should be analyzed by specialized personnel in one of the explorations with the greatest interobserver discrepancy, also considerably increasing their workload.
The development of a semi-automatic tool for the detection of pulmonary nodules could be able to detect early stages of the disease, even when the changes in density level in the lung parenchyma are so subtle that human eyes cannot distinguish. To do this, it should be noted that close collaboration is required between radiologists, given that they know the pathophysiological process of the disease as well as its findings in imaging tests, as well as with the developers of the tool, since they have the knowledge to translate the necessary variables and the development and integration of relevant applications.