# Influence of the number of images on threshold image contrast measurements with a phantom with gold disks in digital mammography

## TL;DR

This study examines how the number of images affects contrast measurements in digital mammography using two versions of a phantom.

## Contribution

The study identifies optimal image numbers for accurate threshold contrast measurements with different CDMAM phantom versions.

## Key findings

- For CDMAM 3.4, statistical differences were found between several image group comparisons.
- For CDMAM 4.0, all groups showed significant differences compared to the N2 group.
- 24 images are recommended for CDMAM 4.0 analysis for reliable results.

## Abstract

Image quality, in addition to radiation dose, is the most important physical parameter in digital mammography. Image quality should be periodically monitored using the CDMAM phantom. The aim of this study is to investigate the effect of the number of analyzed images on the result of threshold image contrast measurements using the CDMAM phantom in different versions. The images obtained using two versions of the CDMAM phantom, i.e., 3.4 and 4.0, were analyzed. The image analysis was performed and repeated 10 times for 2, 4, 6, 8, 12, 16, 24, and 32 images from a pool of 43 images, separately for each phantom. For the CDMAM 3.4 phantom, a statistical difference was demonstrated between the following groups: S2 vs S6 (p < 0.006), S6 vs S16 (p < 0.001), S6 vs S24 (p < 0.002), S6 vs S32 (p < 0.021), S8 vs S16 (p < 0.019), S8 vs S24 (p < 0.048). For the CDMAM 4.0 phantom, a statistically significant difference was demonstrated between all groups and the N2 group (p < 0.000). For the CDMAM 3.4 phantom, the most favorable number of images required for analysis cannot be clearly determined. For the CDMAM 4.0 phantom, it is recommended to perform 24 images for analysis. Particular attention should be paid when determining the threshold image contrast for a disk diameter of 0.1 mm, as this parameter is used during exposure automation control.