Master thesis

Classification performance analysis of medical histopathology images using deep neural networks

💡 This repository contains Jupyter Notebooks and Python scripts used during master thesis research.

• Thesis
• Thesis Presentation
• Conference Paper

Repository structure

.
├── notebooks                   # Jupyter Notebooks
├── etc
│   ├── diagrams                # draw.io diagrams 
│   ├── pseudocode              # LaTeX code for iterative algorithm
│   └── ...
└── README.md

Jupyter Notebooks

1. Data_preview
   • Downloads the datasets
   • Prints the samples from each dataset
2. Data_preprocessing
   • Converting images from .tif into .png
   • Resizing images
   • Datasets merge
3. Training
   • Train and save models for model/dataset combinations
4. Testing
   • Load saved models and evaluate them
   • Store evaluation results for future analyze without loading datasets/models
   • Analyze the evaluation results
5. Iterative V1
   • Single shared test set iterative algorithm implementation
   • Train and analyzis included

⚠️ Some notebooks use the results of the previous ones so they need to be executed in order.

Datasets (Human Colorectal Cancer Histological Images)

Dataset name	URL data	URL paper	Size	Resolution	Classes	Balanced	Stain normalization
Colorectal Histology	Link	kather16	5000	150x150	8	Yes	No
NCT-CRC-HE-100K	Link	kather19	100000	224x224	9	No	Yes
CRC-VAL-HE-7K	Link	kather19	7180	224x224	9	No	Yes

Random samples from datasets

Colorectal Histology

---

NCT-CRC-HE-100K dataset