Mahmoud Shiri - Department of Biomedical Engineering, School of Medicine, Shahid Beheshti University of Medical Sciences
Masoumeh Gity - Department of Radiology, School of Medicine, Tehran University of Medical Sciences
Ali Ameri - Department of Biomedical Engineering, School of Medicine, Shahid Beheshti University of Medical Sciences

Introduction: Breast cancer is the most common type of cancer, and mammography is the main screening test for breast cancer. To assist radiologists in detecting breast cancer from mammograms, computer aided detection (CAD) systems have been developed. Due to recent improvements in software and hardware resources as well as access to larger datasets, there is a growing interest in improving the performance of CAD systems. Methods: This study proposes a deep convolutional neural networks (CNN) for automatic detection of the location of calcifications in mammograms. For this purpose, a ResNet CNN was fine-tuned on ۱۵۴۷ mammographic images, and was then tested on ۳۲۷ images, form the DDSM dataset. Results: The proposed model was applied on ۱۶۰×۱۶۰ patches of each image to identify if it contains calcification. Moreover, the proposed model was tested on ۱۰ mammograms from our in-house dataset. The results showed ۹۱% accuracy in detecting the location of calcifications. Conclusion: These promising results highlight the potential of deep learning in automated detection of breast cancer which can improve CAD systems performance.

Deep learning, convolutional neural network, mammography, breast cancer, CAD, calcification