Quality assessment

Note: We have released MRI data used in this case. (see details in zenodo.org)

[ ]:

!gdown --id 1OjvZB4C9SawoDfMYcpgJsMBsIl4FYnj5

/usr/local/lib/python3.7/dist-packages/gdown/cli.py:131: FutureWarning: Option `--id` was deprecated in version 4.3.1 and will be removed in 5.0. You don't need to pass it anymore to use a file ID.
  category=FutureWarning,
Downloading...
From: https://drive.google.com/uc?id=1OjvZB4C9SawoDfMYcpgJsMBsIl4FYnj5
To: /content/Fig7-uc-individual&pixel-level.xlsx
100% 22.3k/22.3k [00:00<00:00, 25.2MB/s]

[ ]:

import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np

data = pd.read_excel(r'Fig7-uc-individual&pixel-level.xlsx')
data.head()

# Formula: sqrt(Alea) + sqrt(Epis) vs Dice
data['sqrt'] = np.sqrt(data['Alea']) + np.sqrt(data['Epis'])
# norm to 0-1
data['sqrt'] = data['sqrt'] - data['sqrt'].min()
data['sqrt'] = data['sqrt'] * 1.0 / data['sqrt'].max()
sns.lmplot(
    x='sqrt',
    y='Dice',
    data=data,

)
plt.xlabel(xlabel='Uncertainty', fontsize=17)
plt.ylabel(ylabel='Dice', fontsize=17)
plt.show()
../_images/notebooks_5_QA_3_0.png

Conclusion

The estimates of uncertainty at the volume level are negatively correlated with the Dice score (r = -0.75), suggesting that BEN’s uncertainty measure can be considered as an alternative metric for assessing the quality of segmentations at the time of inference, when the Dice score cannot be computed because the ground-truth segmentation is not available.

[ ]: