Dual Multi-Scale CNN for Multi-layer Breast Cancer Classification at Multi-Resolution
Nithin Joseph, Ruchika Gupta · 2022
The resolution dependent textures & diverse color of histopathological images comprising the hard-to-discern high coherency of cancerous cells which makes it very challenging to categorize breast cancer. As such, generic learning models are proposed in the existing state-of-the-art unable to address the different magnification factor with single learning stage. Additionally, the feature extracted from the existing deep learning based models often unable to extract rotation or scale-invariant features which is a key in addressing the breast cancer from rich zoom intrinsic histopathological images. In this work, we have proposed fully-automated robust Dual Multi-Scale CNN capable of extracting linearly seperable rich scale-invariant features. Dual Multi-Scale CNN has been validated on large scale open source BreakHis dataset in different magnification factor. Additionally, our proposed model can be deployed at tertiary hospital settings as clinical deployment in hand-held smart devices for accurate BC categorization.