Development of a Convolutional Neural Network for detection of Lung Cancer based on Computed Tomography Images

Lung cancer is a disease that generates an impact worldwide as it is one of the cancers with the highest incidence and mortality among all types of cancer. Lung cancer usually does not show symptoms until it is in an advanced stage of the cancer, lowering the survival rate of those affected. At the national level, lung cancer is recognized as the second type of cancer with the highest incidence and the first with the highest mortality, so an early and effective diagnosis can increase the survival rate. Currently there are many techniques for the diagnosis of cancer in early stages such as Computed Tomography (CT). CT scans are widely used due to their high sensitivity in the detection of pulmonary nodules without the need to be invasive, which avoids physical damage to the tissues. In addition, there has currently been a great growth in advanced computational techniques that, together with CT, can extract useful characteristics that support radiologists in the detection process. One of the most used techniques for this purpose are neural networks, which are designed to identify patterns and classify image objects. For this reason, in this project the methodology presented consists of training two convolutional neural networks with the VCC16 and DenseNet networks, which have been used in other projects with good results. Both methods were trained with the LIDC-IDRI database, which has a large number of CT scans of patients diagnosed with lung cancer. An evaluation of the models was carried out based on the metrics of accuracy, precision, sensitivity and specificity where the values of 0.992, 0.859, 0.952 and 0.994 were obtained for the VCC.16 network and for the DenseNet network the values 0.995, 0.909, 0.959 and 0.996 were obtained respectively. Demonstrating the relevance of this technique in the medical field, serving as support for radiologists and later serving as an alternative for lung cancer detection.