Hidden

COVID-19 CT Images Segmentation

Description

For COVID-19 patients the common stage in diagnosis is computer tomography (CT). A radiologist is often asked to estimate the extent of damage with respect to lung volume. It is a time-consuming procedure, because radiologist should look through all axial slices on CT and segment each of them.

There are some COVID-specific findings: https://radiologyassistant.nl/chest/covid-19-corads-classification. In this challenge you are asked to segment "ground-glass" and "consolidation".

The goal of this challenge is to help companies, who are building CT-specific software, come up with better solutions.

Metric

Submissions are evaluated on pixel-wise F1-score. F1-score is averaged across all pixels of 10 test images and two classes. Similarly, as it is a classification challenge

File Format

Your submission file should be in csv format, with a header and columns names: Id,Predicted.

If you have numpy array test_masks_prediction with shape (10, 512, 512, 2) - it can be converted to submission the following way:

import pandas as pd

pd.DataFrame(
             data=np.stack((np.arange(len(test_masks_prediction.ravel())), 
                                        test_masks_prediction.ravel().astype(int)),
                                       axis=-1), 
             columns=['Id', 'Expected'])\
.set_index('Id').to_csv('submission.csv')