Authors:
M. E. Fantacci
1
;
S. Bagnasco
2
;
N. Camarlinghi
3
;
E. Fiorina
4
;
E. Lopez Torres
5
;
F. Pennanzio
4
;
c. Peroni
4
;
A. Retico
3
;
M. Saletta
2
;
C. Sottocornola
1
;
A. Traverso
6
and
P. Cerello
2
Affiliations:
1
Pisa University and Pisa Section of INFN, Italy
;
2
Torino Section of INFN, Italy
;
3
Pisa Section of INFN, Italy
;
4
Torino Section of INFN and Torino University, Italy
;
5
Torino Section of INFN and CEADEN, Italy
;
6
Torino Section of INFN and Politecnico di Torino, Italy
Keyword(s):
Computer Aided Detection, Lung Nodules, Thoracic Computed Tomography.
Related
Ontology
Subjects/Areas/Topics:
Bioinformatics
;
Biomedical Engineering
;
Image Analysis
;
Pattern Recognition, Clustering and Classification
Abstract:
M5L, a Web-based fully automated Computer-Aided Detection (CAD) system for the automated detection of lung nodules in thoracic Computed Tomography (CT), is based on a multi-thread analysis with two independent CAD subsystems, the lung Channeler Ant Model (lungCAM) and the Voxel-Based Neural Analysis (VBNA), and on the combination of their results. The lungCAM subsystem is based on a model of the capabilities that ants show in nature in finding structures, defining shapes and acting according with local information. The VBNA subsystem is based on a multi-scale filter for spherical structures in searching internal nodules and on the analysis of the intersections of surface normals in searching pleural nodules. The M5L performance, extensively validated on 1043 CT scans from 3 independent datasets, including the full LIDC/IDRI database, is homogeneous across the databases: the sensitivity is about 0.8 at 6-8 False Positive findings per scan, despite the different annotation criteria and
acquisition and reconstruction conditions. A prototype service based on M5L is hosted on a server operated by INFN in Torino. Preliminary validation tests of the system have recently started in several Italian radiological institutes.
(More)