Evaluation of the Relationship Between CDMAM Contrast Thresholds and an NPWE-Based Detectability Index d′

Aim and Research Question(s)

This thesis evaluated the relationship between CDMAM-derived contrast threshold thickness values (T) and the NPWE-based detectability index (d′) to assess whether d′ could serve as an objective alternative or complement to conventional CDMAM-based image quality assessment in digital mammography. Research questions:

• How does the logarithm of contrast threshold thickness in CDMAM phantom data relate to the logarithm of detectability index d′?
• Can the detectability index d’ replace the CDMAM for image quality assessments in digital radiology systems?

Background

In European screening programs, image quality in digital mammography is commonly assessed using the CDMAM phantom. [3] However, automated CDMAM require multiple image aquisitions and may be influenced by technical factors, which could affect measurement precision and reproducibility. [1] Based on physical image quality metrics such as the modulation transfer function (MTF) and normalized noise power spectrum (NNPS) metrics, the detectability index (d′) can be calculated using a non-prewhitening model observer with eye filter (NPWE). As an objective and observer-independent measure of signal detectability, d′ has been proposed as a potential alternative or complement to conventional CDMAM-based image quality assessment. [2]

Methods

Retrospective CDMAM phantom datasets from 21 mammography systems were provided by the Austrian Agency for Health and Food Safety (AGES). The datasets included automated CDMAM evaluations with threshold thickness values (T). Detectability index values (d′) were calculated using a MATLAB-based graphical user interface implementing a non-prewhitening model observer with eye filter (NPWE), which was based on MTF and NNPS data from the same systems. Logarithmic regression analysis was performed to investigate the relationship between T and d′ for selected CDMAM detail diameters in the CDMAM 3.4 and CDMAM 4.0 datasets.

Results and Discussion

A consistent inverse log-log relationship between threshold thickness (T) and detectability index (d′) was observed across all evaluated datasets. Moderate to strong correlations were found, with R² values ranging from 0.6321 to 0.9038. Higher d′ values were consistently associated with lower threshold thickness values and vice versa. The CDMAM 3.4 datasets showed slightly higher R² values than the CDMAM 4.0 datasets.

Conclusion

The results indicate that d′ could provide a promising objective metric for mammography image quality assessment. However, further validation is required before it can replace conventional CDMAM-based quality assurance.

References

[1] De las Heras Gala, H., Schöfer, F., Tiller, B., Chevalier del Río, M., Zwettler, G., & Semturs, F. (2012). A new method for dosimetry and image quality assurance in mammography and breast tomosynthesis. [2] Monnin, P., Marshall, N. W., Bosmans, H., Bochud, F. O., & Verdun, F. R. (2011). Image quality assessment in digital mammography: Part I. Technical characterization of the systems. Physics in Medicine and Biology, 56(14), 4201–4220. [3] Young, K. C., Cook, J. J. H., Oduko, J. M., & Bosmans, H. (2006). Comparison of software and human observers in reading images of the CDMAM test object to assess digital mammography systems (M. J. Flynn & J. Hsieh, Hrsg.; S. 614206).