Room: Exhibit Hall | Forum 1
Purpose: The early screening and diagnosis of breast cancer are helpful to decrease the morbidity and mortality. In mammography screening, cranio-caudal (CC) and medio-lateral oblique (MLO) projections are normally two mammographic views for each breast. Locating corresponding lesions in two mammographic views of ipsilateral breast is important for radiologist. In this work, we propose corresponding model of different-view mammograms based on back propagation (BP) neural network in digital mammography.
Methods: We constructed the corresponding modelby using BP neural network. Specifically, the proposed method consists six steps: 1) Breast outline extraction and muscle line determination; 2) Nipple coordination computation by ellipse-fitting; 3) Characteristic data acquisition; 4) Straight matching strip detection by BP neural network, which is parallel to muscle line; 5) Arc-shaped matching strip detection by BP neural network, whose circular center is nipple coordinate; 6) Intersection of straight and arc-shape matching strip. To verify the proposed method, we aggregate 316 pairs of mammograms including 565 microcalcifications which could be observed in CC and MLO mammograms and centroid positions of microcalcifications have been determined by radiologists.
Results: Our method has been proved to have better results than conventional straight and arc-shaped matching method without BP neural network. The experiments show that the mean absolute error between the centroid positions of microcalcifications and matching centerlines in straight and arc-shaped matching strip was 3.55mm and 3.51mm, respectively. When the detective sensitivity is 95%, the intersection matching strip was Â±11.06mm which is much smaller than intersection matching strip without BP which was Â±20.51mm.
Conclusion: Compared with conventional matching method, the proposed method decrease matching area and increase detection sensitivity in CC and MLO image. And the experimental results show great potential for lesions matching of mammograms
Funding Support, Disclosures, and Conflict of Interest: National Natural Science Foundation of China (81301940 and 81428019), National Key Research and Development Program (2016YFA0202003), Guangdong Natural Science Foundation of China (2016A030310388 and 2017A030313692), and Southern Medical University Startup fund (LX2016N003).
Mammography, Mammographic Screening, Image Correlation