Generated vascaulature embeddings for retina biomarker discovery

This repository contains all code used for generating the results described in:

Giancardo, L., Roberts, K. & Zhao, Z. Jan 1 2017 Fetal, Infant and Ophthalmic Medical Image Analysis - International Workshop, FIFI 2017 and 4th International Workshop, OMIA 2017 Held in Conjunction with MICCAI 2017, Proceedings.

The network weights file (test6/test6_best_weights.h5) need to be downloaded at the following address: https://drive.google.com/file/d/1UzAVgJKIsVFqQPfg-wk7d4yGW5k1gbCJ/view?usp=sharing and placed in the test6 directory.

The following file shows how to generate the embedding from a test image on the 'data' folder: testEmbedding.py

It also include a visual ouput of the vasculature segmentation used to drive the embedding.

N.B. The function generating the embeddings reloads on the fly the model see the generateEncoding. An example of a more efficient way of generating the encodings for large dataset is in the generateEncoding function of transfLearning.py

These are the packages that need to be installed for Python 2.x

conda install python=2.7 tensorflow-gpu keras pandas scikit-learn scikit-image matplotlib seaborn opencv ipython jupyter configparser

The code was tested using the coda environment in this file: conda_env.yml

it can be replicated by running:

conda create --name environmentName --file conda_env.yml

An initial version of a Python 3-compatible codebase has been added.