Final group project of the signal and image processing course.
https://github.com/signal-and-image-processing-fs19/oct-image-srf-detection.git

The goal is to find a image processing solution to detect from OCT images, if it shows SRF (subretinal fluid) which is a diagnostic factor in detecting degenerative retinal diseases. sub-retinal fluid (SRF) corresponds to the accumulation of a clear or lipid-rich exudate in the sub-retinal space, i.e., between the photoreceptor layer and the underlying retinal pigment epithelium (RPE).

What to look for: homogeneous hypo reflective well-defined areas between the retinal pigment epithelium layer (RPE) and photoreceptor layer.

Textbook example of sub-retinal fluid:

Created on Python version: 3.7.3

Clone the project and run the following command from the root folder to install all dependencies.

pip install -r requirements.txt

Execute 'project_Waelchli_Moser_Meise.py' (wrapper) or 'oct_srf_detection.py':

python project_Waelchli_Moser_Meise.py

Creates (further description see 'Output' below):

• 'project_Waelchli_Moser_Meise.csv': main output with image classification results as specified in 'Test-Data/submission_guidelines.txt'
• 'log.dat': log txt-file of the stdout from running the program
• 'figures/{figname}.png': threshold-optimizing plot

Make sure to install new packages using the following commands in order to make sure that the dependencies are listed in the requirements.txt file:

pip install <package> 
pip freeze > requirements.txt

Training Data consists of srf examples ('Train-Data/SRF/') and non-srf examples ('Train-Data/noSRF/') 15 images each.

Test Data consists of 100 oct images ('Test-Data/handout/) of unknown srf status.

Images are in the png format (rgba) but can be transformed to gray scale after read in, since the data is in gray-scale (0-255).

If a template is specified manually it should also be in the same format.

• 'project_Waelchli_Moser_Meise.csv': main output with image classification results as specified in 'Test-Data/submission_guidelines.txt'
  example_submission.csv:

filename,label
990.png,0
9752.png,1
9703.png,0
8382.png,1
...

• 'log.dat': log txt-file of the stdout from running the program
• 'figures/{figname}.png': threshold-optimizing plot