Skip to content
Segmentation Guided Scoring of Pathological Lesions in Swine Through CNNs
Python
Branch: master
Clone or download

Latest commit

Fetching latest commit…
Cannot retrieve the latest commit at this time.

Files

Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
code
imgs
.gitignore
README.md

README.md

Segmentation Guided Scoring of Pathological Lesions in Swines Through Convolutional Neural Networks

Code

Our UNet-based model can be found in code/model.py.

Our rule-based classifier is available as a single function in code/get_lesion_score_from_segmentation.py

Dataset

Annotation Process

The experts followed a layer-wise annotation strategy. Thus, each anatomical structure is fully annotated even when covered by others:

Example
Layers of the annotation process.

Pleural Lesion Scoring Test Set

Test set with 200 examples (50 for each class) is available with no restrictions HERE

Each example comprises:

  • An RGB image (.jpg) with resolution 1040x780 pixels;
  • A stacked segmentation (.npy) with shape 1040x780xC, where C is the number of segmentation classes

Associated labels for the pleurisy score task are available in the raw_labels.txt file. Examples from the 4 classes are reported:

Example
Class i example (Healthy).

Example
Class ii example (Lesion on Chest Wall 1).

Example
Class iii example (Lesion on Chest Wall 2).

Example
Class iv example (Lesion on both Chest Walls).

Train Set and Pre-Trained Weights

Experts put a lot of their valuable time in the annotation process without direct retribution. As such, we must keep track of the dataset's uses by the community.

Research Purposes

Please drop us an email.

Economic Purposes

Please drop us an email.

You can’t perform that action at this time.