Data Description

Task:

Detection of abnormalities: Can you detect the abnormal regions in a gastroscopic image?

Dataset:

We totally collect about 800 gastroscopic images from 137 volunteers. Three senior experts are   invited to annotate the lesion/abnormal region independently, and we average them to generate   the pixel-level ground truth.

The data is split into a training and test set (see below). 

Training Data

Dataset Image Num Volunteer Num Resolution Ground Truth

Training Data

Link Download

Abnormal Normal Lesion Health 489*409 Pixel-level
260 205 98 71

Test Data

Dataset Image Num Volunteer Num Resolution Ground Truth

Test Data

Link Download

Abnormal Normal Lesion Health 489*409 Pixel-level
129 104 83 62

Reference Standard

For evaluation, we test the performance based on the image-level predictions. For positive images, one image can be considered as true positive if at least 40% of the truly abnormal pixels are detected, otherwise it will be considered as false negative. For negative images, one image can be considered as true negative only when no abnormal pixel is detected, otherwise it will be considered as false positive.   Let

  • TP be the number of true positives,
  • FP be the number of false positives,
  • TN be the number of true negatives,
  • FN be the number of false negatives,

TPR  =   TP / (TP + FN), FPR  =   FP / (FP + TN),       We choice different thresholds and compute the TPR and FPR accordingly to generate the Receiver Operating Characteristic (ROC) curve. Furthermore, the Area Under Curve (AUC) is used for statistical evaluation.