In this section we roughly explain the computational cost of the proposed algorithm. The cost of the global algorithm can be divided into two different blocks: one for training the system and the other one to match the templates.
For training the system the algorithm firstly computes the eigenfaces and their contours for creating the templates and their deformations. On the other hand, it also computes the 2DPCA approach for the false positive reduction model. Both process are relatively fast, taking around one minute in conventional computers (Windows-based P-IV, and programming in Matlab).
On the other hand, the matching process is slower than the
training step. The matching first reduces the mammograms using a
multilevel approach, obtaining a number of possible RoIs.
Subsequently, these RoIs are dealt with using bigger resolution.
Thus, the amount of time needed in this step highly depends on the
number of these suspicious regions. For instance, the mean of all
cases takes about five minutes per image.