Detection of Pneumothorax: Comparison of High-Resolution Workstation Images, Processed Images and Conventional Radiography.

Javier Quiles¹, Pablo G. Tahoces², Miguel Souto¹, María J. Lado¹, Arturo J. Méndez²,

José Correa², Anxo Martínez¹, Manuel García¹, José M. Carreira¹, Juan J. Vidal¹.

¹ Department of Radiology of the University of Santiago de Compostela,

(Complejo Hospitalario Universitario de Santiago)

²Department of Electronics and Computer Science of the University of Santiago de Compostela

Santiago de Compostela, SPAIN

Introduction

An essential element for the succesful conversion from film-based radiology to a total digital radiology system is the image display. However, visualization can be adversely affected by limited spatial resolution and luminance constraints imposed by monitors. Thus, further investigation is needed to formally establish the dignostic accuracy of digital displays [1].

Postprocessing is another important factor in digital radiography. It has been pointed out that unsharp masking, when globally applied, can distort the appearance of pulmonary markings and degrade observer perfomance in the identification of different pathologies. Adaptive techniques can be alternately used to enhance the radiopaque regions of the thorax while providing little or no enhancement in the lung [2].

We compared detection of pneumothorax on a digital workstation with conventional chest images. Conventional radiographs were also compared to digital images processed with a global filtering and with an adaptive filtering.

Material and Methods

a) Hardware and software features,

110 posteroanterior chest radiographs were digitized using a Konica laser film scanner ( 2430 x 2000 pixels; 0.175 mm/pixel) and 1024 gray levels per pixel. Hard copies of the images were produced with a laser printer (Agfa Matrix Compact L). Software for visualization was developed on a Sun Sparc10 UNIX workstation based on the X Window System and Motif libraries. Images were displayed on a 19” video monitor at full resolution.

b) Methodology

Four experienced radiologists examined a total of 440 images (110 for each image modality: conventional, globally filtered, adaptively filtered and non-processed images displayed on monitor). For each set of 110 images, 44 were normal cases, 47 were cases with pneumothorax and 19 were control cases with other abnormalities. Readers indicated the presence or absence of radiographic abnormalities on a 5-point confident scale. We used the negative and positive scores to calculate individual readers sensitivity and specificity. ROC curves were drawn to characterize each radiologists’ perfomance with each of the image modalities.

Results

Sensitivity and specificity values were calculated for each reader and each image modality (Table I). The comparison among the four image modalities was carried out by means of the areas under the ROC curves (Table II). The Friedman non-parametrical test indicated that there were no statistical differences among any of the four modalities for the detection of pneumothorax.

	Table I. Sensitivity and Specificity
		Sensitivity            TPF=           True Positive Responses                                     Total Number of Positive Cases					Specificity TNF=        True Negative Responses                          Total Number of Negative Cases
			Conventional	Monitor	Global Filtering	Adaptive Filtering		Conventional	Monitor	Global Filtering	Adaptive Filtering
	Reader 1		0.87	0.87	0.91	0.96		0.93	0.90	0.98	0.91
	Reader 2		0.81	0.81	0.87	0.89		0.90	0.84	0.93	0.91
	Reader 3		0.79	0.79	0.83	0.87		0.90	0.88	0.98	0.95
	Reader 4		0.81	0.87	0.85	0.91		0.88	0.90	0.98	0.95
	Average		0.82	0.83	0.86	0.91		0.90	0.88	0.97	0.93

Table II. Areas under the ROC curves
	Conventional	Monitor	Global Filtering	Adaptive Filtering
Reader 1	0.97±0.02	0.91±0.05	0.96±0.03	0.95±0.03
Reader 2	0.88±0.04	0.86±0.05	0.95±0.03	0.95±0.02
Reader 3	0.87±0.05	0.89±0.05	0.93±0.04	0.86±0.05
Reader 4	0.89±0.04	0.91±0.04	0.96±0.03	0.96±0.02
Average	0.90±0.04	0.89±0.05	0.95±0.03	0.93±0.03

Discussion

          Global processing, by enhancing pulmonary vasculature, may affect the ability of radiologists to diagnose subtle areas of disease. Adaptive techniques have been designed to avoid undesired enhancement in the lungs; however, in our study there were not significant differences between both global and adaptive algorithms for the detection of pneumothorax.

          Our results suggest that images on a 2430 x 2000 matrix displayed on workstations may be an acceptable alternative to conventional radiographs for the detection of pneumothorax. The use of digital processing techniques for high frequency enhancement, either as a global or as an adaptive filtering, did not significantly improve results from conventional radiographs and workstation images.

References

[1] Brent K. Stewart, Thurman Gillespie, Thomas a. Spraggings, Samuel J. Dwyer III. Functionality of Gray-Scale Display Workstation Hardware and Software in Clinical Radiology. Radiographics 14:657-669. 1994

[2] Correa JL, Souto M, Tahoces PG, Malagari KS, Tucker D, Larkin JJ, Kuhlman J, Barnes GT, Zerhouni EA, Fraser RG, Vidal JJ. Digital chest radiography: comparison of unprocessed and processed images in the detection of solitary pulmonary nodules. Radiology; 195: 253-258. 1994

Corresponding Author

Javier Quiles del Río

Department of Radiology, University of Santiago de Compostela

Facultad de Medicina

C/ San Francisco 1

15705 Santiago de Compostela

SPAIN

tel 34 81 570982

fax: 34 81 547031

e-mail: mrquiles(at)uscmail.usc.es

---

Oral presentation at EuroPACS'98, Barcelona, Spain

---