Using these image quality criteria, for serial studies (such as in screening), it is advantageous to review previous images, when available, prior to imaging the client. This practice allows previous areas of difficulty to be identified (e.g. thin pectoral muscle or lack of infra mammary angle); furthermore comparison between current and previous mammograms enables the observer to check whether the entire breast has been included. Detailed information on client positioning can be found in Chaps.​ 21, 22 and 23.

Compression force should be sufficient to separate the overlying structures in the breast, to create a uniform and reduced tissue thickness and to immobilise the breast – thereby minimising the potential of motion unsharpness [5, 6]. The reduced tissue thickness minimises geometric unsharpness and scatter, both of which should enhance image quality. Further discussion on compression force application can be found in Chap.​ 22.

Exposure factors normally determined by the imaging equipment. These are optimised to enable detail in both dense glandular and less dense fatty tissues to be demonstrated, breast tissue to be seen through the pectoral muscle on the MLO projection and the skin edge to be visualised.

See Chap.​ 16 for further information on exposure factors.

As developing cancers can have similar density to glandular breast tissue, high contrast is essential to differentiate suspicious features from normal appearances. There are many variables which affect contrast between lesions and surrounding tissue, these include exposure factor optimisation, use of compression force and breast position. Good technique is therefore necessary for adequate lesion visibility.

In the clinical setting, sharpness is related to displaying distinct anatomical features with clear edges. Lack of sharpness increases the risk of low density lesions being missed and some features being incorrectly characterised. The sharpness of an image is related to all of the following: [7, 8]

Noise gives the image a grainy, mottled appearance and can obscure or even mimic small lesions. If noise is present then the perception of microcalcifications can be challenging. Further information about noise can be found in Chap.​ 16.

An assumption of the ability to detect features representing pathology in radiology is that it is related to image quality – if quality increases then pathology detection should, generally, increase too. Assessment of quality by visual means is clinically realistic and if done adequately it will have valuable implications for the imaging service. However, assessing image quality by visual means can be hard to achieve, if the assessment is to give accurate and repeatable results and if it is going to predict diagnostic performance.

Radiology and radiography literature is plagued with poorly designed and poorly implemented methods of visually assessing image quality. For many imaging procedures European quality criteria highlighting specific anatomical structures have been defined and these are often referred to for research and clinical purposes. Attempts have been made to update and translate these into visual grading criteria suitable for digital imaging. Unfortunately the original European quality criteria can only give an assessment of how well an image will perform for very general clinical tasks and they may be inadequate at predicting diagnostic performance for specific pathologies. Mammography is no different to the rest of radiology and radiography, since more rigorously validated image quality criteria which can be more task specific do not exist.

As we have already seen, within mammography, various clinically important anatomical structures for the mammogram have been identified that carry information concerning the presence of pathology; the ability to visualise these structures is used as a basis for visual image quality assessment. The important underlying assumption with this is the detection of pathology correlates well with the visibility of this normal anatomy.

Building on the criteria, visual grading analysis tools (eg PGMI and EAR) have been created and they remain in common use. The use of such tools, it was hoped, would minimise subjectivity and also offer the potential to provide a numeric value of visual image quality that would correlate with cancer detection. However, similar to the criteria, none of the visual grading analysis tools used in mammography have been validated. For clinical and research purposes there is a need for robust visual grading scales to be created and validated. Below we explain one approach on how this might be achieved.

Bandura’s [14] theory provides a suitable theoretical basis for visual grading scale development and validation [15, 16]. This is because visual image quality evaluation requires interaction between human attitude/perception and physical attributes in an image. Psychometrics, a branch of psychology, deals with measuring human attributes that cannot be measured directly. In this context, the [psychometric] visual grading scale would comprise a set of statements (items) that attempt to measure perception of visual image quality in a valid and consistent fashion. Using Bandura’s theory, visual grading scale development and validation comprises several steps.

First, a draft set of quality statements (scale items) is created using generic [17] and mammography specific literature. They would include essential visual anatomical characteristics that should reflect mammographic image quality. Second, a focus group of clinical mammography experts would review and, if required, modify the items. The items are then worded positively (50 %) and negatively (50 %), to minimise affirmation bias. Then, a Likert scale is included for scoring. A Likert scale of 1–5 is suitable, where 1 would be strongly disagree with the item; 5 would be strongly agree with the item. Second, a set of approximately 7 FFDM mammographic image sets, with qualities varying from poor to excellent, are identified through consensus by a panel of experts. Physical measures, such as signal to noise ratio, could be included to assist the selection process. Third, the draft scale is pilot tested with a small number of clinical mammography professionals to identify and correct any ambiguity associated with item wording. Fourth, the scale is used to assess visual quality of the 7 mammographic image sets by suitably trained clinical mammography professionals. To reduce error, at least 150 professionals should do this, resulting in 7 × 150 completed visual grading scales. Fifth, the data should be analysed statistically to validate the visual grading scale [18, 19]. This analysis can result in several scales being produced, examples could include: 1. Full scale to assess left or right CC/MLO image sets; 2. Full scale for CC only and full scale for MLO only; 2. Shorter scales (with fewer scale items) for ‘1’ and ‘2’.