As number of breast cancer patients are increasing, the cost and time required for diagnosis can be minimized using computer-aided pathology to analyze microscopic histopathology images. Some deep learning approaches outperforms the task of classification but they lack in interpretability. This model provides better interpretability along with classification result by providing localization on microscopic histopathology images. In this approach, an image is cropped into small square patches and a bag is made out of it. The bag of images will act like a batch. This bag is passed to convolutional neural network for feature extraction. Then instance level features are passed to classifier to get the instance level attention (in terms of weight). At the end an image is stiched from cropped patches by multiplying with attention weight resulting in classified image with highlighting useful portion. The datasets used are BreakHIS and BACH.
Input variables : Microscopic histopathology image
Output Variables : Classified image (benign/ malignant) with important portion highlighted
Visit Model : github.com
Additional links : arxiv.org
| Model Category | : | Public |
| Date Published | : | April, 2020 |
| Healthcare Domain | : | Provider |
| Code | : | github.com |
| Health Risk Management |
| Disease Detection |