(Sunday, 7/14/2019)
Room: ePoster Forums
Purpose: Computer-aided diagnosis has been investigated over 30 yrs. Deep learning (DL) has been embodied into novel frameworks for mammographic breast cancer diagnosis (MBCD). This study aims to provide a technique review of DL based MBCD in a systematic manner.
Methods: IEEEXplore, Pubmed, ScienceDirect and Google Scholar have been used to search literatures relating to DL based MBCD. The last update was at December 20, 2018. Keywords include “convolutional neural network�, “deep learning�, “mammography�, “breast cancer� and “diagnosis�. Specifically, only papers published on peer-reviewed journals were selected and our search yielded 18 research articles. To each publication, the databases, the number of breast lesions and the best performance are reported. Furthermore, the models are analyzed from technical details to the pros and cons.
Results: The DL-based MBCD models are categorized into three groups: 1) dedicated DL models – designing shallow or modifies networks to decrease the time cost as well as the number of instances for training; 2) transferred DL models – making the best use of pre-trained CNNs by transfer learning and parameter tuning; and 3) information fusion models – taking advantage of CNN models for feature extraction, while the differentiation of malignant lesions from benign ones is fulfilled by machine learning classifiers. Due to limited public datasets, there are 7 articles evaluated on in-house collection and 12 articles on less than 1000 medical instances. Achieved AUC ranges from 0.72 to 0.98 and ACC from 0.81 to 0.98.
Conclusion: This review presents the recent progress of DL based MBCD. It is found that DL can enhance MBCD performance. While due to few large-scale databases, full exploration and comparison of DL-based MBCD models remains challenging.
Funding Support, Disclosures, and Conflict of Interest: This work is partly supported by grants from National Key Research and Develop Program of China (2016YFC0105102), Leading Talent of Special Support Project in Guangdong (2016TX03R139), Fundamental Research Program of Shenzhen (JCYJ20170413162458312), the Science Foundation of Guangdong (2017B020229002, 2015B020233011, 2014A030312006) and the National Natural Science Foundation of China (61871374).
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
