Low Computational Model for Classifying Medical Histology Images.
Abstract
Breast Cancer is one of the most common types of cancer amongst women, and early detection of carcinogenic tissue from histology images can go a long way in effective treatment as well. Recent approaches to solving this problem of classification of tissues utilize heavy deep learning-based architectures which takes higher inference time and storage space for model parameters. In this paper, we propose a CNN architecture with a lesser number of parameters that can be effectively deployed on a resource-constrained device, with the utilization of the Knowledge Distillation technique. We evaluate our approach on High-Resolution Breast Cancer histology slides of the BACH 2018 dataset.