Display of mammograms on a CRT

Transcription

1 Display of mammograms on a CRT Hans Roehrig, Ph.D. William J. Dallas, Ph.D. Elizabeth Krupinski, Ph.D. Jiahua Fan, M.S. University of Arizona This work was supported by 2 Grants from NIH

2 In most radiological imaging workstations the monochrome cathode ray tube (CRT) is the electronic display of choice. The CRT offers the best performance; it is the most highly developed, mature and reliable display in common use. Unfortunately, because of limited spatial resolution and because of limited dynamic range, the efficacy of soft-copy versus hard-copy diagnosis has been challenged.

3

4 It has been shown that, when the images are displayspecifically processed before the display to compensate for the reduced MTF of the CRT and its reduced dynamic range, current high-resolution monochrome CRT display systems will present all the information in mammograms without interaction by the reader 1,2. The basic idea is to perform an operation on the image before it is displayed that will counteract the degradation that will come from the action of the display system s MTF.

5 This paper discusses display specific processing with respect to spatial resolution (MTF) for the specific case of Mammography. The display specific processing is based on the evaluation of the CRT image quality with the aid of a CCD camera.

6 The CCD Camera was used to image test-stimuli like squarewave patterns and single lines to determine the MTF It was also used to take images of some of the processed and unprocessed images as they appeared on the CRT for the purpose of close inspection and analysis.

7 We use the Wiener-Helstrom filtering method to prepare the image for display. Denoting the original image u(x,y), its Fourier transform is U(ξ,η). Modeling the display as an approximately linear shift-invariant system, the output will be degraded on the display by the optical transfer function OTF(ξ,η) according to the equation below, where N represents the Fourier spectral density of the noise. V ( ξ, η) = U ( ξ, η)[ OTF( ξ, η) + N( ξ, η)] The use of the Wiener filter in this form comes from the fact that the actual restoration we are attempting is variations in the point spread function, not in the image.

8 Because the OTF is positive over the entire region of frequency space, we may replace it by its absolute value, the modulation transfer function (MTF). In actual practice, we replace the noise power spectral density by a constant that we adjust as a parameter in the filter. We make an additional simplification and assume the MTF to be separable so that we can actually perform the measurements and the filtering as 1-D operations. MTFξ ( ξ ) MTFη ( η) F( ξ, η) = MTF ( ξ ) N MTF ( η) N ξ + Caveat: There is subtle ringing in the processed image resulting from discontinuities in the unprocessed image, especially at the borders. The MTF of the CRT reduces ringing. Total removal of ringing requires extra work. ξ η + η

9 Processing and ringing Image with single line: unprocessed Image with single line: processed Ringing Profile of the above unprocessed single line image Profile of the above processed single line image with ringing

10 Processing and ringing (Continued) Single line image: unprocessed and displayed on the CRT Single line image: processed and displayed on the CRT Ringing is reduced because of the CRT MTF Profile of the above unprocessed single line image Profile of the above processed single line image with subtle ringing

11 MTFs(from the Square Wave response) of a5m-pixel Monitor with a P45 Phosphor for Different Mean Luminances (Nyquist Frequency: 3.47 lp/mm) Vertical MTFs cd/m 2 (ADU 55) Compensated Horizontal MTFs of a 5M-pixel Monitor with a P45 Phosphor for Different Estimates(n) of Noise (Luminance: 64 cd/m 2, Nyquist Frequency: 3.47 lp/mm) MTF cd/m 2 (ADU 55) Horizontal MTFs cd/m 2 (120 ADU) cd/m 2 (120 ADU) cd/m 2 (ADU 210) cd/m 2 (ADU 210) Spatial Frequency(lp/mm) MTF Unprocessed Vertical MTF 1 2 Unprocessed Horizontal MTF Spatial Frequency (lp/mm) 12 Note that n is a parameter for noise we adjust in the Wiener filter. The larger n, the larger the estimate of the noise, and the lower the boost on the high frequency.

12 Typical Mammogram Right MLO Right CC Microcalcifications, which need to be displayed on a CRT with high fidelity.

13 Pixel Magnitude (CCD Camera ADU) Profiles of a Micro-Calcification before and after processing with n = 1 to compensate for the monitor MTF The slopes of the Micro_Calcification are noticeably steeper after the processing.the Micro-Calcification appears now much sharper. Processed Unprocessed Pixel Magnitude (CCD Camera ADU) Profiles of a Micro-Calcification after processing with n = 1 and n = 10 The slopes of the Micro_Calcification after processing with n = 1 are slightly steeper than after processing with n = 10.The Micro-Calcification appears slightly more sharp for n = 1 Processed with n = 1 Processed with n = Pixel Position Pixel Position

14 Examples of Processing of Microcalcifications Unprocessed image from the database Unprocessed image as displayed on CRT Processed image as displayed on CRT Difference between processed image and unprocessed image

15 Profiles of the calcification in the center of the previous images Pixel Magnitude (CCD Camera ADU) Profiles of a Micro-Calcification before and after processing with n = 1 to compensate for the monitor MTF The slopes of the Micro_Calcification are noticeably steeper after the processing.the Micro-Calcification appears now much sharper. Processed Unprocessed Pixel Magnitude (CCD Camera ADU) Profiles of a Micro-Calcification after processing with n = 1 and n = 10 The slopes of the Micro_Calcification after processing with n = 1 are slightly steeper than after processing with n = 10.The Micro-Calcification appears slightly more sharp for n = 1 Processed with n = 1 Processed with n = Pixel Position Pixel Position

16 Example of Processing of a Benign Microcalcifications Unprocessed image from the database Unprocessed image as displayed on the CRT Image on the film Processed image as displayed on the CRT Difference between processed image and unprocessed image

17 Results of a Pilot ROC Study Mean ROC Az % 50% 75% 100% Overall Unprocessed Processed Lesion Contrast Performance as measured by ROC Az (Receiver Operating Characteristic Area Under the Curve) was higher at all contrast levels for the images processed for the monitor MTF compared to the unprocessed versions, but the difference was not significant.

18 The data are based on 2 observers viewing sets of 250 images in both the unprocessed and MTF-correction processed images. Lesion contrasts are at 25%, 50%, 75% and 100% (original images). Performance as measured by ROC Az (Receiver Operating Characteristic Area Under the Curve) was higher at all contrast levels for the MTF-correction processed images compared to the unprocessed versions. With only 2 observers at this time, none of the conditions reached statistical significance, although the early trends suggest that with more readers some of the contrast level conditions (50%, 75% and overall) are likely to reach statistical significance.

19 Conclusion Image processing by using the Wiener-Helstrom filtering method compensates for the blurring caused by the MTF of the CRT. Initial results indicate that this compensation is clinically useful. Display specific processing must also include consideration for the reduced dynamic range, which is not included in this study but will be implemented in the future.

20 References 1. Reiker GG, Blume H, Slone RM, Woodard PM, Gierada DS, Sagel SS, Jost RG, Blaine GJ: Film less Digital Chest Radiography within the Radiology Department; SPIE Vol 3035, , Roehrig H, The Monochrome Cathode Ray Tube Display and its Performance ; Volume 3, Chapter 4 of Yongmin Kim and Steven C. Horii, eds, Display and PACS ; in Jacob Beutel, Harold L. Kundel and Richard L. Van Metter, eds, Handbook of Medical Imaging, SPIE Press, Bellingham,WA, June 2000.