Exploring Histological Similarities Across Cancers From a Deep Learning Perspective

This code is for our paper titled Exploring Histological Similarities Across Cancers From a Deep Learning Perspective published at Frontiers in Oncology, Vol 12, 2022.

Authors: Piyush Singh*, Ashish Menon*, C. V. Jawahar, P. K. Vinod

📝 Paper 📑 Demo Page 📑 Code for demo
paper website demo code

Summary: In this work, we trained 11 patch classifiers for the Cancer vs Normal task on 11 different cancer types. Then we perform cross inference. Further we use each of these classifiers to generate RoIs using GradCAM and measure overlap with respct to the model trained on each cancer type. Furthermore, we study the similarities in the histograms of geometric features within these RoIs to enhance this understanding even more.

Architecture

Patch CNN architecture

arch_v2_lite(1)

Nucleus geometry analysis workflow

nuc_seg_wflow

Results

Cross Inference Results

grid_v2

GradCAM Overlap

gradcam_v2(1)

Nucleus geometry analysis

BRCA_KICH_COAD_combined(1)

License and Citation

The software is licensed under the MIT License. Please cite the following paper if you have used this code:

@article{Menon2022ExploringHS,
  title={Exploring Histological Similarities Across Cancers From a Deep Learning Perspective},
  author={A Vipin Menon and Piyush Singh and P. K. Vinod and C.V. Jawahar},
  journal={Frontiers in Oncology},
  year={2022},
  volume={12}
}