Automated registration of diagnostic to prediagnostic x-ray mammograms: Evaluation and comparison to radiologists’ accuracy
Example of simulated mammograms from magnetic resonance breast images.
Example of landmarks identified by a mammography film reader and an affine registration. Footnote: Identified points on the diagnostic film (left, white) and the mean locations of the landmarks identified by the same reader on the prediagnostic film (right, white) plus landmark locations generated from the affine registration algorithm (right, black). (Points 1 and 2 indicate the tumor, points 3 and 4 indicate normal features, and point 5 indicates the nipple.)
Prediagnostic image showing the two sets of points identified by a reader at two separate sittings ( and , grey and black points) and the corresponding points from the affine registration (, white points).
Parameter combinations used to train the affine, FFD, and fluid registration methods.
Parameter combinations for the affine, FFD, and fluid registration methods which produced the most accurate registrations for the simulation training set of 20 registration pairs (ten women).
Registration accuracy for mammogram simulation test set of 20 image pairs created from ten women’s MRIs.a
Characteristics of the study subjects and of their tumors. density; deviation, and range.
Mean within-reader and between-reader error distances. interval.
Population mean (95% CI) for registration errors by landmark and breast density (data adjusted for radiologists working conditions). interval; density.
Descriptive statistics for the registration methods by landmark and breast density. density.
Article metrics loading...
Full text loading...