A Clinically Viable Approach to Lung Segmentation and Nodules Reinclusion

The World Health Organization has estimated 14.1 million new cases of cancer in 2012 and 8.8 million deaths in 2015. Lung cancer being the most fatal and prevalent of all. Therefore it is important for public health to use means that provide faster and more accurate diagnoses. Computer Assisted Diagnosis systems (CADs) have the potential to improve the accuracy of imaging diagnoses and are effective in compensating for the deficiency in the performance of the human eye in the detection of minor lesions. Mainly, when lung segmentation fails to delineate the lungs correctly, nodules may be lost. This work proposes an approach with 3D connected components analysis for lung segmentation. The results shows very little loss in the number of nodules lost with only 1.9% of 2663 nodules present in the dataset, in an average time of 19.3 s per exam, compared to 223 s of the 3D region growing method.