1. Goal: 使用 YoloV8 对医学影像CT/磁共振图像进行肺炎区域分割:

part1 :

A. init data [nii(Nifit) -> png]: 将CT原始数据从 三维转二维, 保存png格式
B. mask data [nii -> json]: 将CT的分割标注数据从 三维转二维, 再转Labelme可以识别的json文件

part2:

Algorithm: Yolo-V8
A. make dataset [train/val/test + label]: 对二维数据和json文件进行数据划分, 并json转换Yolo标签格式
B. Segmentation [Yolo-V8]: 使用YoloV8进行分割

2. JSON format (eg:) :

{
"version": "4.5.6", // Labelme software version
"flags": {}, // Additional flag information
"shapes":
{
"label": "dog", // Label name
"points": [ // Polygon vertex points of the annotation
[298, 151,
[324, 151],
[324, 160],
[317, 160],
[317, 168],
[307, 168],
[307, 151],
[298, 151]
],
"group_id": null,
"shape_type": "polygon", // Shape type
"flags": {}
},
{
"label": "cat",
"points":
[400, 200,
[425, 200],
[425, 210],
[415, 210],
[415, 220],
[405, 220],
[405, 200],
[400, 200]
],
"group_id": null,
"shape_type": "polygon",
"flags": {}
}
...
],
"imagePath": "path/to/image.jpg", // Image file path
"imageData": null, // Image data (Base64 encoded)
"imageHeight": 480, // Image height
"imageWidth": 640 // Image width
}

YOLO .txt format:
one object -> one row
(classID, x_center, y_center, width_normalized, height_normalized)

(eg):
1 0.375000 0.625000 0.312500 0.416667
0 0.656250 0.291667 0.156250 0.208333

dataset.yaml format(eg):
train: /path/to/your/dataset/images/train
val: /path/to/your/dataset/images/val
nc: 5
names: ['cat', 'dog', 'person', 'car', 'truck']