OS: Ubuntu 18.04
Required software
python 3.7.7
tensorflow 2.2.0
tensorflow-addons 0.10.0
For a complete list of python packages see requirements.txt file
Installation of python dependencies
pip install -r requirements.txt
Installation of the echoAI-PET-measurements module into the virtual environment (e.g. conda, virtualenv or pipenv)
python setup.py develop
Now the main function that runs the model can be imported
import numpy as np
from echoai_pet_measurements.runCFRModel import runCFRModel
An echocardiography video can be loaded into the memory as a numpy array
# Load video data into memory
with open(datafile, 'rb') as f:
data = np.load(f)
# Provide frame time (in ms) and image scaling factors
frame_time_ms = 30
deltaX = 0.03667
deltaY = 0.03367
Inference on the echocardiography video by calling the runCFRModel function
_, predictions = runCFRModel(data_array_list=[data],
frame_time_ms_list=[frame_time_ms],
deltaX_list=[deltaX],
deltaY_list=[deltaY])
Estimated time for running the code on a single echo input: 5.4 seconds.
If no weights are available, a new model with random weights is initialized. However, checkpoint files can be provided to the runCFRModel function in the form of a python dictionary, where
checkpoint_dict = {'model_response_variable': 'path_to_checkpoint_file'}
Moreover, multiple videos can be provided to this function as lists. The returned qualified_index_list contains the video indices of those videos that satisfied the minimum requirements (maximum frame_time and minimum length).
qualified_index_list, predictions = runCFRModel(data_array_list,
frame_time_ms_list,
deltaX_list,
deltaY_list,
checkpoint_dict,
batch_size=1)