Keywords:
Breast, Ultrasound, Screening, Diagnostic procedure, Localisation, Cancer, Image registration
Authors:
M. Horn1, D. Kim2, G. Ng3, B. N. Bloch1; 1Boston, MA/US, 2Toronto/CA, 3Bothell/US
DOI:
10.1594/ecr2018/C-2794
Results
The AI Breast proved to offer a reliable tool to accurately assess the location of the transducer and underlying target,
without a significant compromise in time for the initial registration.
The median time taken to perform the initial registration was 12 seconds,
across the study group.
Timing was heavily dependent on operator experience.
The electromagnetically tracked positional data was used to apply automatic annotations during the clinical exam.
Each clinical finding was annotated with laterality,
clock position,
and distance from the nipple (Figures 4,
5,
& 6).
Each of the automatic annotations applied were confirmed by more than one operator in each clinical session,
while a match between the physical location of the transducer on the breast and the virtual display was observed.
The real-time dynamic display of the transducer graphic streamlined clinical work flow by eliminating the need to make the manual adjustments for each documented image (Figures 3,
4,
5,
& 6).
With significant reduction of user interaction,
especially when multiple lesions were present,
less time per patient exam was noted.
The automated localization and correct annotation in AI Breast also improved reproducibility when imaging identified lesions for follow up examinations.
There was increased confidence on follow up,
particularly when there was a need to differentiate between several similar looking lesions in a given area.
Further,
mislabeling was completely avoided (i.e.
wrong breast: left verus right; mix-up of clock positions: 3 vs 9 o'clock or 11 vs 1 o'clock).