CheXBox : Lungs detection in chest radiographs¶

Chest X-ray is the most commonly acquired image in medicine. Chest X-ray uses a very small dose of ionizing radiation to produce pictures of the inside of the chest. It is used to evaluate the lungs, heart and chest wall and may be used to help diagnose shortness of breath, persistent cough, fever, chest pain or injury. It also may be used to help diagnose and monitor treatment for a variety of lung conditions such as pneumonia, emphysema, and cancer. Because chest X-ray is fast and easy, it is particularly useful in emergency diagnosis and treatment.

Because of the difference in density between air, soft tissue and bone, the lungs appear much darker than their surroundings. Brighter regions in the lungs may indicate the presence of pathology.

The data used in this assignment belong to the Chestxray14 dataset, which is publicly available and can be found at this link:https://nihcc.app.box.com/v/ChestXray-NIHCC. The dataset, released by the NIH, contains 112,120 frontal-view x-ray images of 30,805 unique patients, annotated with up to 14 different thoracic pathology labels using NLP methods on radiology reports.

For this challenge, we have selected 13,331 (1000 for training, 3,331 for testing) chest X-ray images from chestxray14 and generated bounding boxes containing the left lung, the right lung. The goal of this challenge is to detect the lungs in chest x-rays. The algorithms should produce the coordinates of the two bounding boxes (for right and left lungs) given a chest x-ray.

Submission Details:¶

Your submission csv should look like this:

  filename,RL_xmin,RL_ymin,RL_xmax,RL_ymax,LL_xmin,LL_ymin,LL_xmax,LL_ymax
  00000013_011,114,25,249,329,318,401,484,401,318,401,484,401
  00000013_038,78,79,263,415,322,81,486,423,322,81,486,423
  ...
  00000475_000,80,37,233,397,286,43,440,367,322,81,486,423
  00000491_009,102,61,237,383,300,53,426,343,322,81,486,423