Keywords:
Computer applications, Computer Applications-General
Authors:
D. Cascio, V. Chietri, F. Fauci, M. Iacomi, R. Ienzi, R. Magro, G. Raso, S. Sorce, M. Vasile Simone; Palermo/IT
DOI:
10.1594/ecr2010/C-1095
Results
The acquired informations by the simulation process have been used to extract data by the mammographic images. In particular, a selected ROIp (p=potentially pathological) in one of the pair projections (RCC/LCC, RMLO/LMLO) have been compared with the exactly symmetrical areas in the other projection of the same pair, obtained after an scale elastic transformation, resulting in projections overlapping of the same pair.
Through the choice of a correct Fcutoff, determined in the simulation process, it is possible a comparison between ROIpr (pr=pathological randomized) and ROIs (s=symmetrical) spectra, in order to eliminate false positives maintaining only the ROI visible in one pair's projection. This is sufficient to ensure the presence of pathology (Fig 1).
In particular, the method evaluates the ROIp elimination when the R value in ROIpr is less than the R value in the ROIs.
The correspondences detection is the most delicate step of the whole process and is determinant to the success of the whole method; it is a subjective activity whose results are liable of errors. This implies the necessity to consider tolerances when identifying a correspondence among the CC and the MLO views is requested. It is reasonable to consider that one or more columns of the MLO view (i.e. left) corresponds to more columns of the other MLO (i.e. right) image.
The method was applied on a dataset composed of 200 couples of pathological images.
In particular, experimental evidence shows that when the patient has pathological mass lesions in only one of the view (i.e. left), the symmetrical view (i.e. right) include only noise, so the signal disappear. Comparing the correlated ROIs with the radiologist diagnosis, we have observed a full matching up to 95% of the cases.