Cancer is the second deadliest disease in the world. Breast cancer tops the
list among the diseases affecting women. Specific strategies should be devised which
will mitigate the effects of breast cancer. The risks can be mitigated if the detection
takes place at an early stage. Early detection leads to improved outcomes, and survival
remains a cornerstone of cancer control. Currently, mammograms are used to capture
and observe the 2D nature of the tissues. 2D mammogram reports are used to train
convolutional neural networks. 2D mammograms capture anterior and posterior images
of the breast. These images, alone, are not sufficient to adjudicate whether the lump is
benign or malign. Convolutional Neural Networks have attained great success in image
classification, but they fail in some areas since they learn about the image statically.
They do not take into consideration spatial information about the image and its subparts. There is no significant change reflected in the output if there is some alteration in
the input. CNNs tend to lose lots of valuable information in the process of pooling. To
overcome all these shortcomings, 3D data will be used to train the network, which
captures all the orientations of the tissues. 3D mammograms, also known as
tomosynthesis, are also very helpful for women who have concentrated dense tissues.
Dense tissues make it difficult to locate the abnormalities. In addition to 3D data,
clinical history, genomic information, and pathology reports have been taken into
consideration. The amalgamation of the heterogenic data helps in the accuracy of the
prediction because it will analyze all the contexts before arriving at a decision. Capsule
neural networks have been used to overcome the drawbacks of convolutional neural
networks. Convolutional neural networks require a lot of training data, which is not
readily available. It takes a lot of time to train the model since the volume of data is
huge. It is not capable of recognizing deformed objects in various orientations. Capsule
Neural Network addresses all these issues and improves the performance reasonably.
Keywords: Capsule neural networks, Convolutional neural networks, Ductal carcinoma, Tomosynthesis.
