Introduction: The purpose of medical imaging in breast cancer screening is to detect and characterise pathology. In this context, image quality is best defined in relation to diagnostic performance. However, in many situations, the only practical and economical method to assess image quality is to use clinical image quality assessment performed by radiologists. Clinical image quality assessment can be expensive, time-consuming and suffers from intra and inter-observer variability. Therefore, it is useful to establish a robust, quantitative image assessment method that can accurately predict radiologists’ clinical image quality assessment. Unfortunately, the variable anatomical backgrounds in clinical images significantly complicate the relationship between physical and clinical image qual- ity. Two-dimensional digital X-ray mammography (DM) is currently the most commonly utilised screening modality in breast cancer screening. However, it has well-documented limitations. Promising alternatives to two-dimensional DM involving phase-contrast X-ray imaging are currently under investigation. Phase-contrast imaging is an Xray imaging technique where image contrast is not only related to the X-ray attenuation properties of tissue (as is the case with conventional radiography) but also the refractive properties of the tissue. One technique currently under active research at synchrotron facilities is propagation-based phase-contrast computed tomography (PB-CT). In this work, we take advantage of the relatively simple anatomical background present in synchrotron-based, thin slice PB-CT images of breasts to formulate a simple, robust relative image assessment model.

Methods: The experimental data analysed in this study included PB-CT scans, which were obtained for twelve whole, intact mastectomy samples at Imaging and Medical beam- line (IMBL) of the Australian Synchrotron. Eleven radiologists assessed overall clinical image quality. Physical image metrics, including contrast, signal to noise ratio, and spatial resolution, were calculated using two different methods for all PB-CT and conventional CT image sets. Weighting factors were applied to each metric, and a scaled contrast to noise (CNR) to spatial resolution (res) score (CNR/res) was calculated.

Results: The scaled CNR/res score for each imaging condition, averaged across all samples, was found to correlate significantly with the corresponding radiologist scores with a Pearson r value of approximately 0.96. In addition, the CNR/res score for each imaging condition, for each sample, also correlated significantly with the corresponding radiologist scores with a Spearman r value of approximately 0.89.

Conclusion: The scaled CNR/res criterion has been demonstrated as a quantitative image assessment model that effectively predicts the relative clinical image quality, as assessed by radiologists, in the context of PB-CT breast imaging.