The purpose of this work is twofold: (i) to develop a CAD system for the assessment of emphysema by digital chest radiography and (ii) to test it against CT imaging. The system is based on the analysis of the shape of lung silhouette as imaged in standard chest examination. Postero-anterior and lateral views are processed to extract the contours of the lung fields automatically. Subsequently, the shape of lung silhouettes is described by polyline approximation and the computed feature-set processed by a neural network to estimate the probability of emphysema. Images of radiographic studies from 225 patients were collected and properly annotated to build an experimental dataset named EMPH. Each patient had undergone a standard two-views chest radiography and CT for diagnostic purposes. In addition, the images (247) from JSRT dataset were used to evaluate lung segmentation in postero-anterior view. System performances were assessed by: (i) analyzing the quality of the automatic segmentation of the lung silhouette against manual tracing and (ii) measuring the capabilities of emphysema recognition. As to step i, on JSRT dataset, we obtained overlap percentage (?) 92.7±3.3%, Dice Similarity Coefficient (DSC) 95.5±3.7% and average contour distance (ACD) 1.73±0.87 mm. On EMPH dataset we had ? = 93.1±2.9%, DSC = 96.1±3.5% and ACD = 1.62±0.92 mm, for the postero-anterior view, while we had ? = 94.5± 4.6%, DSC = 91.0±6.3% and ACD = 2.22±0.86 mm, for the lateral view. As to step ii, accuracy of emphysema recognition was 95.4%, with sensitivity and specificity 94.5% and 96.1% respectively. According to experimental results our system allows reliable and inexpensive recognition of emphysema on digital chest radiography.

A computer-aided diagnosis approach for emphysema recognition in chest radiography.

Coppini G;Miniati M;Monti S;Paterni M;Favilla R;Ferdeghini E
2013

Abstract

The purpose of this work is twofold: (i) to develop a CAD system for the assessment of emphysema by digital chest radiography and (ii) to test it against CT imaging. The system is based on the analysis of the shape of lung silhouette as imaged in standard chest examination. Postero-anterior and lateral views are processed to extract the contours of the lung fields automatically. Subsequently, the shape of lung silhouettes is described by polyline approximation and the computed feature-set processed by a neural network to estimate the probability of emphysema. Images of radiographic studies from 225 patients were collected and properly annotated to build an experimental dataset named EMPH. Each patient had undergone a standard two-views chest radiography and CT for diagnostic purposes. In addition, the images (247) from JSRT dataset were used to evaluate lung segmentation in postero-anterior view. System performances were assessed by: (i) analyzing the quality of the automatic segmentation of the lung silhouette against manual tracing and (ii) measuring the capabilities of emphysema recognition. As to step i, on JSRT dataset, we obtained overlap percentage (?) 92.7±3.3%, Dice Similarity Coefficient (DSC) 95.5±3.7% and average contour distance (ACD) 1.73±0.87 mm. On EMPH dataset we had ? = 93.1±2.9%, DSC = 96.1±3.5% and ACD = 1.62±0.92 mm, for the postero-anterior view, while we had ? = 94.5± 4.6%, DSC = 91.0±6.3% and ACD = 2.22±0.86 mm, for the lateral view. As to step ii, accuracy of emphysema recognition was 95.4%, with sensitivity and specificity 94.5% and 96.1% respectively. According to experimental results our system allows reliable and inexpensive recognition of emphysema on digital chest radiography.
2013
Istituto di Fisiologia Clinica - IFC
Chest radiography
Emphysema
Computer-aided-diagnosis (CAD) systems
Neural networks
Lung segmentation
File in questo prodotto:
File Dimensione Formato  
prod_209748-doc_47815.pdf

accesso aperto

Descrizione: A computer-aided diagnosis approach for emphysema recognition in chest radiography.
Dimensione 1.59 MB
Formato Adobe PDF
1.59 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.14243/174830
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 37
  • ???jsp.display-item.citation.isi??? ND
social impact