Computational assessment of mammography accreditation phantom images and correlation with human observer analysis

Automated image quality assessment of mammography phantoms: a systematic review

“…The phantoms used in these studies were commercially available for the evaluation of mammographic image quality. In DM studies, four studies (7,12–14) used the American College of Radiology (ACR) mammography accreditation phantom, two studies (3,10) the TORMAS phantom, four studies (16,18–20) the CDMAM phantom, and two studies (17,20) used the CIRS SP01 phantom. One study (17) used an in-house designed phantom named the RACON phantom, and another study (15) developed software to analyze various quality control measures from several modalities (X-ray radiography, mammography, angiograms, and fluoroscopy).…”

“…The image acquisition parameters in different studies varied as shown in Tables 2 and 3; AEC, variable mode of mAs (tube current and exposure time product), kVp (tube potentials), detector technology and anode/filter arrangement. Similarly, the image quality metrices used in these studies were different and included: automated visibility scoring method (5–9,12,14,20–23,25,26,28), contrast (4,13,16–18,27), mean pixel value (13), contrast-to-noise ratio (CNR) (12,13), signal-to-noise ratio (SNR) (4,17,18,24,27), contrast-detail scoring (16,18,19,29), resolution as line pairs per millimeter (lp/mm) (17,20,25), modulation transfer function (MTF) (18), image quality figure (IQF) (16,17,20), and image quality index (IQI) (3,10,20,25). Although most studies selected only a specific object to assess in the phantom, 13 out of 26 studies assessed all the test object features available in the studied phantom.…”

“…In this review, we noted that some studies evaluated the software’s performance compared to observers, while other studies only validated the software based on the information from the manufacturer. In addition, only three studies (7,10,14) validated the software over the long term (one year (14), two years (10), and three years (7)), whereas other included studies did not use it over the long term. We also observed that the original ACR mammography accreditation phantom was the most commonly used phantom in the experiments.…”

“…Such variations in human assessment make tracking changes in the mammography system difficult, and this could mask any failings in image quality and performance of the system. In contrast, methods of computerized image analysis provide more consistent and reproducible results over the long term as demonstrated in the studies by Barufaldi et al, (14) Gennaro et al (10), and Sundell et al (7), which were carried out using image data acquired over one-, two-, and three-year periods, respectively. Therefore, they were able to detect image quality deviations and underlying issues that might have impacted the quality of the digital mammography systems' images (3,7,10,14).…”

“…Despite all selected studies involving a quantitative method to analyze the quality of the mammography phantom images, they varied in materials and image quality parameters as indicated in Tables 2 and 3. The studies included were separated into two groups based on the image acquisition type: 12 studies used DM for image acquisition (3,7,10,(12)(13)(14)(15)(16)(17)(18)(19)(20), while 14 studies used (digitized) SFM images (4)(5)(6)8,9,(21)(22)(23)(24)(25)(26)(27)(28)(29).…”