@author Rahul Dhodapkar rahul.dhodapkar@yale.edu ORCID: 0000-0002-2014-7515 @version 2022.07.03
This project contains code to prototype the automated quality control of en-face images for optical coherence tomography-angiography (OCT-A).
To use the pre-trained high quality and low quality models, simply load the fastai models
from their respective pickle files into a python environment.
For a sample bitmap image at ./testimg.bmp, one could run:
#####
# load libraries
#
from fastai.vision.all import PILImage
from fastai.learner import load_learner
#####
# load models
#
low_quality_model = load_learner('./calc/fastai/hisens_model.pkl')
high_quality_model = load_learner('./calc/fastai/hispec_model.pkl')
#####
# load image
#
img = PILImage.create('./testimg.bmp')
#####
# run loaded models
#
lq_is_valid, _, lq_probs = low_quality_model.predict(img)
hq_is_valid, _, hq_probs = high_quality_model.predict(img)
The relative confidence of each model is encoded in the returned objects from the .predict() call.
All data and model files are available upon request from the authors jay (dot) wang (at) yale (dot) edu or rahul (dot) dhodapkar (at) yale (dot) edu.