Automated System for the Detection of Lung Nodules
Fecha
2021-11
Autores
Martinez Machado, Elizabeth
Perez Diaz, Marlen
Orozco Morales, Rubén
Título de la revista
ISSN de la revista
Título del volumen
Editor
Resumen
Lung cancer is the most frequent cause of cancer mortality in the world. The diagnostic
procedure usually begins with a chest X-ray; however, it is difficult to interpret due to
the set of anatomical structures overlapped. Computer-aided detection (CAD) systems
are a diagnostic aid tool for radiologists. In the present work a CAD system is proposed
for the detection of lung nodules on chest radiographs. Methods such as convolution,
local normalization and homomorphic filters are used to pre-process images, using a
multi-level threshold method supported by morphological operations for anatomical
segmentation. This is followed by a candidate nodule detector using the local slidingband
convergence filter. The candidate nodules are segmented using an adaptive
threshold based on distance. A set of characteristics for each candidate are calculated
based on the segmentation. The system was tested by a free available database (DB) of
247 images, of which 154 are pulmonary nodules (100 malignant and 54 benign cases
and 93 nodules). The results obtained indicate that the system is able of detecting
98.7% of the nodules of the DB with an average of 56.08 detections per image. Two false
positive were obtained due to lung segmentation.
Descripción
Palabras clave
CAD System, Lung nodules, Image processing