Beautiful does not mean better: Clinically-relevant assessment of mammographic image quality
Digital mammography (DM) is the predominant examination used for breast cancer screening. When reviewing the mammogram, image quality plays an essential role in lesion-detection performance. Image quality can be affected by acquisition parameters and image processing. To guarantee optimal performance of DM systems, guidelines have been developed to assess, optimise, and approve DM systems before they are introduced into clinical practice or screening. An example is type testing, in which physical and clinical assessments are performed. In the latter, image quality issues that affect image interpretation are considered but not evaluated with respect to the clinical task. This means that this assessment tends to result in a “beauty contest”, i.e., issues not relevant to detection or characterisation of lesions, but associated, for instance, with aesthetics, can lead to high scores. However, a beautiful image does not mean a better image. Therefore, the goal of this thesis was to develop, evaluate, and validate an instrument for clinically-relevant image quality assessment.
In Chapter 1, a previously-developed algorithm showed satisfactory accuracy in simulating different technique factors. It was validated using realistic mammograms produced from 3D-printed breast phantoms of varying thickness and breast composition. The algorithm was used to simulate image quality issues throughout this thesis. In Chapter 2, the effects of low image quality on radiologists’ perceived ability to interpret a mammogram were investigated. Results showed that interpretation was affected by acquisition-related issues as well as post-processing-related issues. Some of those issues affected calcification cases more than soft tissue cases. Chapter 3 describes a mixed-method study in which qualitative and quantitative methods were used to develop and evaluate an instrument to assess mammographic image quality. It resulted in 18 items of clinically-relevant structures of normal and abnormal tissue. In Chapter 4, the clinical relevance of the instrument was investigated against receiver operating characteristics analysis, the gold standard for assessing clinical performance, and for different professionals. Moreover, it was investigated whether the instrument is successful and reliable in assessing image quality. This study showed a potential correlation between the image quality assessment using the instrument and clinical performance, which means it may be used to assess clinically-relevant image quality.
Overall, this thesis presents an instrument that underwent development, evaluation, and validation processes, illustrating the value of using clinically-relevant structures related to breast tissue and lesions to clinically assess mammographic image quality, and of using psychometric theory in developing instruments to assess image quality.