Lung-Nodule Classification Based on Computed Tomography Using Taxonomic Diversity Indexes and an SVM

• Published in: Journal of signal processing systems Vol. 87; no. 2; pp. 179 - 196
• Main Authors: de Carvalho Filho, Antonio Oseas, Silva, Aristófanes Corrêa, de Paiva, Anselmo Cardoso, Nunes, Rodolfo Acatauassú, Gattass, Marcelo
• Format: Journal Article
• Language: English
• Published: New York Springer US 01-05-2017; Springer Nature B.V
• Subjects: Circuits and Systems; Computed tomography; Computer Imaging; Electrical Engineering; Engineering; Image Processing and Computer Vision; Pattern Recognition; Pattern Recognition and Graphics; Signal,Image and Speech Processing; Vision; Medical image; Phylogenetic trees; Taxonomic diversity index; Taxonomic distinctness; Lung cancer
• ISSN: 1939-8018; 1939-8115
• DOI: 10.1007/s11265-016-1134-5

The present work aims to develop a methodology for classifying lung nodules using the LIDC-IDRI image database. The proposed methodology is based on image-processing and pattern-recognition techniques. To describe the texture of nodule and non-nodule candidates, we use the Taxonomic Diversity and Taxonomic Distinctness Indexes from ecology. The calculation of these indexes is based on phylogenetic trees, which, in this work, are applied to the candidate characterization. Finally, we apply a Support Vector Machine (SVM) as a classifier. In the testing stage, we used 833 exams from the LIDC-IDRI image database. To apply the methodology, we divided the complete database into two groups for training and testing. We used training and testing partitions of 20/80 %, 40/60 %, 60/40 %, and 80/20 %. The division was repeated five times at random. The presented methodology shows promising results for classifying nodules and non-nodules, presenting a mean accuracy of 98.11 %. Lung cancer presents the highest mortality rate and has one of the lowest survival rates after diagnosis. Therefore, the earlier the diagnosis, the higher the chances of a cure for the patient. In addition, the more information available to the specialist, the more precise the diagnosis will be. The methodology proposed here contributes to this.