Convolutional Neural Networks to Analyse Varyingly Stained Rhabdomyosarcoma Histology Images

This page refers to a project to use Convolutional Neural Networks (CNNs) to analyse varyingly stained images of Rhabdomyosarcoma (RMS). This Repo documents implementations, outputs and project files surrounding this project.

RMS is one of the most common extracrannial tumours affecting adolescents. The tumour is difficult to classify and treat. The tumour is primarily classified using histopathology, particularly through Hematoxylin and Eosin (H&E) staining. This project icnorporates four stain methods, they are:

• H&E
• Trichrome
• MYoD1
• Myogenin

This project leverages multiple Neural Network advancements and techniques, including the following methods:

1. Data Augmentation
2. Stain Normalisation
3. Transfer Learning
4. CNN Model Merging
5. Post Training Visualisation

We use existing Github-hosted libraries Stain Normalization and LIME.