<title>Method To Study The Effect Of Controlled Changes Of Breast Image Total System Modulation Transfer Function (MTF) On Diagnostic Accuracy</title>

Abstract

AbstractOne hundred and fifty conventional craniocaudad mammographic images representing various benign and malignant breast conditions were obtained and four separate copies of the original images were made, each having a specific and different change in resolution. Three experienced mammographers evaluated the randomly arranged original images and copies and recorded their analysis. The analysis included visibility of various anatomical breast structures and pathological lesions (masses and calcifications), subjective image quality, and mammographic interpretation. The resulting 2,250 evaluations were analyzed by Receiver Operating Characteristics Analysis. No statistically significant difference in performance was demonstrated by the radiologists in the range of the resolution changes evaluated.IntroductionA general problem in diagnostic radiology is to relate the physical factors affecting image quality to the diagnostic image quality as measured by diagnostic accuracy.1 Despite the various techniques available, relating the physical descriptors of the imaging process and the empirical descriptors of observer performance remains a major problem area.Numerous physical factors affecting image quality can be measured such as film contrast, noise power spectrum, and unsharpness. (Modulation Transfer Function (MTF)). Observer performance may be described by Receiver Operating Characteristic (ROC) curve analysis and the McNemar test.The ROC curve analysis has been used to compare different imaging modalities, such as CT nuclear medicine imaging-^ and screen/film radiography.10The purpose of this study was to vary only the spatial resolution in mammographic images and analyze the diagnostic accuracy of experienced mammographers for the different quality mammographic images.Materials and methodsOne hundred and fifty Kodak Min-R screen/film combination craniocaudad mammographic images were selected from the files of the Department of Diagnostic Radiology at M.D. Anderson Hospital and Tumor Institute, Houston, Texas. The mammograms included 88 cases with various benign diseases and 62 cases with histologically verified malignant disease. The average patient age was 55 years old. Of 70 images, with visible calcifications, 30 had calcifications less than 0.5iran in size. Forty-six images demonstrated irregular border­ ed masses greater than 1cm in size. Smooth bordered masses were visualized on 30 images, and 20 of these exceeded 1cm in size.Controlled changes in resolution of the images were made while maintaining film contrast and noise constant. Several methods for changing the resolution of the radiographs including optical filtering, digital filtering, and printing techniques were investigated. For this study a printing technique^ was selected based on equipment availability, cost, technical complexity, reliability, and time.The original mammographic image was printed on copy film (DuPont Duplicating Film) using a DuPont Cronex printer. Changes in resolution were obtained by inserting different numbers of sheets of clear base film (Kodak clear base film NMC-1, O.D. = 0.02) between the original and the copy film. Four separate copies of the original were made: the contact copy (emulsion to emulsion), a reversed contact (emulsion-film-emulsion), one layer of NMC-1, arid three layers of NMC-1. In total, 750 images resulted (600 copies, 150 originals).Film contrast was evaluated by the characteristic curve of the Min-R screen/film combina­ tion obtained by inverse square sensitometry. Film density strips were used to evaluate the reproducibility of the contrast in the copying provess. In Figure 1, the density of the