Histopathological cancer detection in whole slide images
Cancer Region of interest Extraction and Machine learning
Aim
Aim of this work is to train a deep learning model with mobilenet backend that can run on any device to identify cancer regions in high quality Whole Slide Images (WSI).
Approach
- Whole slide images are first split into patches (smaller region) of size 512 * 512 with stride as 96, then each of this patches is fed into trained Mnasnet (Platform-Aware Neural Architecture Search for Mobile) model.
- Mnasnet then classifies all the patches in the WSI into two classes. A positive label indicates that there is atleast one pixel of tumor tissue.
- Threshold parameter is used to select regions of interest as bounding boxes cordinates. For example if threshold is set to 0.5, the bounding boxes of regions with more than 50% probability of having tumor tissues is returned.
Dataset
- This dataset contains 2,20,025 examples of two classes (positive and negative). A positive label indicates that the center 32x32px region of a patch contains at least one pixel of tumor tissue.
Work in progress
- Converting all this into browser based model using Tensorflow js
- Trying out object detection based modeling instead of Classification (I think it is bit complicated because of the dimensionality of WSI in training the model. On resizing or cropping WSI images we might loose important details like nuclei count)
Future Work
- Train Mnasnet on ICIAR2018 dataset where we have four classes: normal, benign, in situ carcinoma and invasive carcinoma. By doing this we can also add one more parameter which can let the user to select the class and set the threshold for that class. Eg., Return the regions where there is 50% likelihood of having invasive carsinoma.
References
- Multiclass Classification of Breast Cancer in Whole slide Images (https://github.com/scottykwok/bach2018/blob/master/Multiclass%20Classification%20of%20Breast%20Cancer%20in%20Whole%20slide%20Images.pdf)
- Two-Stage Convolutional Neural Network for Breast Cancer Histology Image Classification. ICIAR 2018 Grand Challenge on BreAst Cancer Histology images (BACH) (https://github.com/ImagingLab/ICIAR2018)