Research aim: Breast cancer is an important global health problem. It represents the most common type of cancer in women with increasing incidence over the last decades. Mammography, a two-dimensional x-ray image of the breast, plays a key role in cancer detection in screening programs. In two-dimensional examination technique overlapping tissue occasionally obscures the view of developing cancer, so research are being conducted to verifying usefulness of a three-dimensional imaging technique called tomosynthesis. The key difficulty in tomosynthesis is poor identification of microcalcifications as breast cancer precursors. The aim of the master’s thesis was to quantify the detection performance of microcalcifications various size in tomosynthesis at different levels of exposure of projection images of the breast.
Methods and materials: Breast projection images used in the master’s thesis were obtained from the TCIA online database of medical images and were a part of the VICTRE study. The reconstruction program of the research program group LAVI from the University of São Paulo was used. It was validated to the breast projection images from the VICTRE study. A model of virtual addition of microcalcifications was developed to study the impact of microcalcifications parameters (location in the breast, size and type of microcalcifications) and scanner settings (exposure time) on imaging performance. The detectability of the microcalcifications on reconstructed breast images was quantified using SNR (signal-to-noise) and CNR (contrast-to-noise) ratio metrics.
Results: The master’s thesis shows the dependences of SNR and CNR ratios on size of type I and type II microcalcifications at different levels of exposure of breast projection images. The visibility of type I microcalcifications increases with size of microcalcifications, while the visibility of type II microcalcifications does not change significantly with respect to size. The general visibility of microcalcifications is highly dependent on exposure of breast projection images and it decreases with decreasing exposure of projection images. The improvement of the CNR ratio of microcalcifications at higher levels of exposure of breast projection images could not have been confirmed due to the shortcomings of the model of virtual exposure of projection images.
Conclusion: The results show how the detectability of the differently sized type I and type II microcalcifications depend on the selection of the threshold frequency value of the filter in breast image reconstruction and on the level of exposure of the projection images. The results indicate the optimal parameters selection for the optimal visibility of microcalcifications on reconstructed breast images. A model of virtual addition of microcalcifications to breast projection images was developed and it has a great potential for further use in the research on reconstruction algorithms and DBT systems.
|
