Skip to content

Latest commit

 

History

History
112 lines (85 loc) · 5.9 KB

m2caiseg.md

File metadata and controls

112 lines (85 loc) · 5.9 KB

m2caiseg

Dataset Information

The m2caiSeg dataset is designed for segmentation of endoscopic images during surgical procedures. It is derived from videos 1 and 2 of the MICCAI 2016 Surgical Tool Detection dataset, from which a total of 307 images were sampled and annotated in more detail at the pixel level. The official dataset is divided into 245 training images and 62 test images. The images in the dataset cover a variety of categories and subcategories, including organs such as the liver, gallbladder, top wall, intestines; surgical tools such as clips, bipolar, hooks, scissors, trimmers; as well as fluids like bile and blood. Additionally, there are two special category labels, unknown and black, used to deal with areas that are obscured by certain tools in some images.

In the field of medical AI, especially in the automation of laparoscopic surgeries, although existing surgical robot technologies are advanced, they still rely on a high degree of expertise and skill. For further automation of surgeries, precise identification and annotation of structures and tools in surgical videos are particularly crucial. Overall, the m2caiSeg dataset provides researchers with a rich and well-annotated resource for analyzing and identifying various structures and tools during surgeries.

Dataset Meta Information

Dimensions Modality Task Type Anatomical Structures Anatomical Area Number of Categories Data Volume File Format
2D Endoscopy Segmentation abdominal cavity abdominal cavity 19 307 jpg, png

Resolution Details

Dataset Statistics size
min 596x334
median 659x369
max 774x434

Label Information Statistics

Category Detection Count Detection Rate Max Area Size Min Area Size Median Area Size
Black 307 100% 99911 18982 75503
Liver 304 99% 241975 3 42413
Fat 297 97% 280020 0 24213
Upperwall 291 95% 209966 8 20638
Hook 289 94% 66113 7 20198
Gall-bladder 284 93% 121786 0 13400
Grasper 279 91% 118227 0 5265
Clip 225 73% 1222 0 53
Trocars 222 72% 283572 0 93
Irrigator 220 72% 26583 0 64.5
Artery 193 63% 17207 22 67
Scissors 181 59% 19432 0 17
Unknown 180 59% 300177 0 7
Specimen-bag 157 51% 54583 0 113
Clipper 123 40% 32521 0 119
Bipolar 106 35% 27244 0 3.5
Intestine 80 26% 292067 269 8631
Blood 24 8% 26279 7 8707.5
Bile 18 6% 25838 0 10

Visualization

File Structure

The official file structure is as follows, including the test and train directories and their subdirectories images and groundtruth, as well as the files within these subdirectories.

m2caiSeg dataset/
├── test
│   ├── images
│   │   ├── 0.jpg
│   │   ├── 475.jpg
│   │   └── ...
│   └── groundtruth
│       ├── 0_gt.png
│       ├── 475_gt.png
│       └── ...
└── train
    ├── images
    │   ├── 00.jpg
    │   ├── 225.jpg
    │   └── ...
    └── groundtruth
        ├── 00_gt.png
        ├── 225_gt.png
        └── ...

Authors and Institutions

Salman Maqbool (National University of Sciences and Technology, Pakistan)

Source Information

Official Website: https://www.kaggle.com/datasets/salmanmaq/m2caiseg

Download Link: https://www.kaggle.com/datasets/salmanmaq/m2caiseg/download?datasetVersionNumber=1

Article Address: https://arxiv.org/abs/2008.10134

Publication Date: 2020-12

Citation

@article{maqbool2020m2caiseg,
  title={m2caiSeg: Semantic Segmentation of Laparoscopic Images using Convolutional Neural Networks},
  author={Maqbool, Salman and Riaz, Aqsa and Sajid, Hasan and Hasan, Osman},
  journal={arXiv preprint arXiv:2008.10134},
  year={2020}
}

Original introduction article is here.