An Level Set Evolution Morphology Based Segmentation of Lung Nodules and False Nodule Elimination by 3D Centroid Shift and Frequency Domain DC Constant Analysis
| Autor: | Umamaheswari Rengasamy, Ganesh Narasimhan, Senthilkumar Krishnamurthy |
|---|---|
| Rok vydání: | 2016 |
| Předmět: |
Lung
Computer Networks and Communications business.industry Centroid Nodule (medicine) medicine.disease 030218 nuclear medicine & medical imaging 03 medical and health sciences 0302 clinical medicine Level set medicine.anatomical_structure 030220 oncology & carcinogenesis Frequency domain medicine Segmentation Computer vision Artificial intelligence medicine.symptom Nuclear medicine business Constant (mathematics) Software Mathematics Calcification |
| Zdroj: | International Journal of u- and e- Service, Science and Technology. 9:187-198 |
| ISSN: | 2005-4246 |
| Popis: | A Level Set Evolution with Morphology (LSEM) based segmentation algorithm is proposed in this work to segment all the possible lung nodules from a series of CT scan images. All the segmented nodule candidates were not cancerous in nature. Initially the vessels and calcifications were also segmented as nodule candidates. The structural feature analysis was carried out to remove the vessels. The nodules with more centroid shift in the consecutive slices were eliminated since malignant nodule’s resultant position did not usually deviate. The calcifications were eliminated by frequency domain analysis. DC constant of nodule candidates were computed in frequency domain. The nodule candidates with high DC constant value could be the calcifications as the calcification patterns were homogeneous in nature. This algorithm was applied on a database of 40 patient cases with 58 malignant nodules. The algorithms proposed in this paper precisely detected 55 malignant nodules and failed to detect 3 with a sensitivity of 95%. Further, this algorithm correctly eliminated 778 tissue clusters that were initially segmented as nodules, however, 79 non-malignant tissue clusters were detected as malignant nodules. Therefore, the false positive of this algorithm was 1.98 per patient. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|