This poster is published under an
open license. Please read the
disclaimer for further details.
Keywords:
Radiation physics, Breast, Mammography, Observer performance, Screening, Physics
Authors:
E. Salvagnini1, H. Bosmans1, C. Van Ongeval1, A. van Steen1, K. Michielsen1, L. Cockmartin1, L. Struelens2, N. Marshall1; 1Leuven/BE, 2Mol/BE
DOI:
10.1594/ecr2015/C-1175
Methods and materials
Simulated microcalcification clusters and masses were inserted into unprocessed mammograms,
processed and then scored by four radiologists.
Images were visualized using the ViewDEX software [3,4].
The radiologists performed a free search study and rated their confidence of the detection of a lesion using a five-point rating scale.
Afterwards,
a JAFROC analysis was applied [5].
Screening patient data,
acquired on two identical digital mammography systems,
were collected during one year and the images were divided into 4 thickness groups,
defined as ‘T-groups’ (T1≤29mm; T2=30-49mm; T3=50-69mm; T4≥70mm).
For each T-group,
130 cranio-caudal lesion-free images were randomly selected - the inclusion criterion was absence of any lesion.
BIRADS density scores and Volpara density maps were obtained for all selected images.
BIRADS density scores were given by the radiologist while the Volpara density map were obtained with the VolparaDensity software (VolparaSolutions,
Wellington,
New Zealand).
Simulated microcalcification clusters and masses were inserted into half of the images with one lesion per image.
A detailed description of the microcalcification cluster and mass models used in this work and how these models were created,
is described in the work of Shaheen et al.
[6,7]. Each lesion was included in each of the four thickness groups; insertions were made into images with the same BIRADS score and at a location with the same Volpara density - crucial for controlling the influence of breast thickness on lesion detectability.
An example of four mammograms selected for insertion is shown in figure 1.
These mammograms present a BIRADS density score of 2 and each of them has a local Volpara glandularity area of 21%.
Figure 2 shows the zoom of the area (mammograms in figure 1) where the lesion was inserted before (upper line) and after (bottom line) insertion,
compressed breast thickness increases from left to right.
An identical lesion template was used for each of the 4 images.