New approaches in breast cancer healthcare based on hi-tech artificial intelligence

Background: Now a day in early stage of breast cancer, mammography is most sensitivetechnique for detection and diagnosis. World health organization (WHO) is proposedmammography for reduce the breast cancer mortality. computer aided detection (CAD) hasimproved radiologists’ ability to detect cancers at very early stages. However, recent advancesin machine learning and data availability paves the way for further improvement of CADsystems. For this purpose, we have developed a deep learning-based method for early detectionof breast cancer from mammograms.Material and Methods: A model based on faster R-CNNs (Region proposal networks andConvolutional Neural Networks) is proposed to identify suspicious lesions including masses andcalcifications, by drawing a bounding box around them. The system then classifies these lesionsas benign or malignant. Furthermore, we have leveraged preprocessing techniques to performcontrast enhancement and histogram equalization, which significantly improved the performanceof our model. The Digital Database for Screening Mammography (DDSM) containingmammograms from 2620 individuals were employed to train the model. Subsequently, thenetwork was fine-tuned using data collected from 20 subjects at a local hospital.Results: The proposed scheme was validated using the INbreast database that includes mammogramsfrom 115 persons. The model achieved an AUC (Area Under the ROC Curve) of 0.96, and outperformsprevious methods tested on the INbreast dataset.Conclusions: The strength of the proposed approach highlights the high efficacy of deep learningbasedmethods for automatic detection of malignant lesions in mammograms. The optimized state-ofthe-art architecture along with the effective preprocessing techniques resulted in the higher performanceof the proposed model compared to that of previous methods. Nevertheless, higher quality and amountof training data can substantially enhance the success of deep learning systems. In future work, we willfocus on collecting additional data to further increase the accuracy of the proposed scheme.

Keywords:

Mammogram , Deep learning , CAD , breast cancer , convolutional neural network