In order to minimise mammographic dose we have developed a multi-faceted strategy utilising big data methodologies. Specifically this comprises of:
- A women specific measure of mammographic dose taking into consideration the glandular composition of the breast.
- A big data analytics engine which computes the dose with a consistent methodology for each and every women imaged throughout an entire screening programme,
and facilitates the detailed comparison of the dose characteristics across all x-ray gantries as functions of both compressed breast thickness,
and breast density.
The first step utilises the VolparaDose (Volpara Solutions,
Wellington NZ) software. This incorporates VolparaDensity (Volpara Solutions,
Wellington NZ) volumetric breast density measurement software to estimate the glandularity of each breast from the mammogram. This is taken as an input to the mean glandular dose (MGD) calculation method presented by Dance et al [4-8],
to yield a women-specific dose value.
The Dance method employs a series of multiplicative factors derived from extensive Monte Carlo simulations that are applied to a measure of the incident air kerma upon the upper surface of the breast (the “entrance dose”). In the case of mammography,
three factors are used:
- The incident air kerma to MGD conversion factor - this depends on beam quality (measured via the Half Value Layer (HVL)) and the compressed breast thickness;
- A correction for any difference in breast composition from 50% glandularity – this also depends on the beam quality measured via HVL and compressed breast thickness;
- A correction for different anode or target materials (which naturally generate different x-ray spectrums).
A further factor is included in the case of tomosynthesis MGD to take consideration of the angle of the x-ray beam.
The VolparaDensity software quantifies the proportion of the breast occupied by fibroglandular tissue by assessing the difference between the incident photon fluence and that measured by the detector (the output),
in terms of the attenuation coefficients of adipose and fibroglandular tissue when scatter is taken into account. This is used to inform the glandularity correction applied in the Dance calculation. A dose specific to the individual women’s glandular tissue proportion within the breast is thus calculated. VolparaDose accepts local physicist calibration measurements for the tube output and half-value layer,
in order to provide the highest accuracy in its MGD values.
The use of VolparaDose has a further benefit in addition to providing a women specific dose. It provides a technique for calculating a universally comparable dose value since it adopts a standardised methodology,
as opposed to the values reported by the modality manufacturers in the DICOM headers,
which will likely differ in their calculations algorithms. The ability to have fully comparable dose values is imperative to the proposed dose minimisation scheme,
since it underlies the use of big data methodology in the Analytics engine: one must compare like with like.
The second step utilises VolparaEnterprise (Volpara Solutions,
Wellington NZ),
an Analytics engine built specifically for large scale quality assurance in mammography screening programmes. The backend is fully DICOM integrated,
and is designed such that every modality within the breast imaging network feeds all the images it acquires into the system for analysis. Thus the big data methodology may be fully exploited with maximum statistical power. The software runs on the Cloud,
which overcomes any potential issues with geographical diversity. The VolparaDose software is used to calculate the MGD of every image fed into VolparaEnterprise.
The front end of VolparaEnterprise is a secure web based dashboard accessible by anybody with the necessary security credentials. This presents an assortment of graphical analysis of dose statistics across the whole programme,
allowing the quality control radiation protection physicist centralised monitoring of doses (for example Figure 1),
together with beam quality and AEC configuration details to provide actionable intelligence to swiftly resolve issues they may identify (Figure 2).
Fig. 1: Real time box and whisker plots of patient dose distribution between units for ranges of compressed breast thickness.
In its simplest form this framework may be used to ensure existing diagnostic reference levels are adhered to,
for example the 3.5mGy for patients with a compressed breast thickness between 50-60mm used in the UK. It may also be used to set new levels at regularly periodic intervals,
according to the observed performance of the current equipment in the programme. However,
in order to utilise the true potential of big data analysis,
rather than comparing to absolute thresholds,
relative dose comparisons may be made between the various x-ray gantries. More specifically,
outlier analysis may be used to identify those units that are in some way performing differently. To use an example following a pattern that has been observed during development of this software,
if 90% of the x-ray units are delivering an average dose to 50-60mm BIRADS C and D density breasts of 1.75mGy,
why are the remaining 10% delivering 2.5mGy? Using traditional dose auditing techniques which simply compare to an absolute threshold,
for example the UK DLR of 3.5mGy,
this anomaly would be overlooked. However,
if correlating this information with cancer detection rates,
finds no benefit to the 43% increase in dose from 1.75 to 2.5mGy,
then the ALARA principal dictates an issue has arisen which should be resolved. The beam quality selection and AEC parameter monitoring shown in Figure 2,
then provides the necessary data at a glance to identify the difference between these units,
for example the use of a softer beam.
Fig. 2: Real time monitoring of beam quality selection, and AEC parameters across the screening programme.