ML project: fracture detection

This is a project for the course Machine Learning. The purpose of this project is to detect fractures in chest bones.

Method

1. Vertebra Localization.
2. Bounding Box Movement Prediction
3. Chest Detection
4. Fracture Detection
5. Post Process

Dataset

To train YoloDetector:

https://disk.pku.edu.cn:443/link/35A835D1CA7A36B07E6C9A17A99F1F89

Additional Annotation

files:

• Additional_anno_train.json
• Additional_anno_val.json

{
  "bbox":
  {
    ...
  },
  "poly":
  {
    ...
  }
}

Boundbox of chest

Item: "bbox".

Each one is a key-value pair.

Key: picture id.

Value: [[left, top],[width, height]]

Outline and ribs

Item: "poly".

Each on is a key-value pair.

Key: picture id.

Value: a list of polylines.

[(x1,y1),(x2,y2),...,(xn,yn)]

The first polyline is the outline of chest bones. Marked clockwise, from bottom-left to top to bottom-right.

The other polylines is ribs. Each one of them started from the outline to spine. Arranged clockwise, from bottom-left to top to bottom-right.

Median data files

spine.json

{
  id:[
    [
      [xc, yc],
      [xtl, ytl],
      [xtr, ytr],
      [xbl, ybl],
      [xbr, ybr]
    ],
    [
      ...
    ],
    ...
  ],
  ...
}

The json is a Dict.

Key: picture id

Value: a list of spine bones.

Each spine bones is a list containing 5 points, which are: the center, the top-left, the top-right, the bottom-left, the bottom-right.

Each point has the format [x, y]

ribs.json

{
  id:[
    [
      [x1, y1],
      [x2, y2],
      ...
    ],
    ...
  ],
  ...
}

The json is a dict.

Key: picture id

Value: a list of rib polylines.

Each rib polyline is a list of points, arranged from spine to the outside.

Each point has the format [x,y].