Kaggle gastrointestinal image segmentation competition baseline

recently ,Kaggle Released UW-Madison Gastrointestinal image segmentation competition . Ask contestants to ：

Based on MRI images of actual cancer patients , Propose a model using deep learning , Automatic segmentation MRI scans Image of the intestines and stomach . The deadline for the game is 2022 year 7 month 14 Japan . Reward up to 2.5W dollar .

This is a CV Direction image segmentation task , Small partners interested in this field , Don't miss it .

Deep eye competition Instructor Frank The teacher provided the of the game baseline.

01

PART

Game Analysis +baseline

1、 Title Link

https://www.kaggle.com/competitions/uw-madison-gi-tract-image-segmentation/overview/description

2、 Title Description

The MRI scan of this competition （MRI scans） Images from actual cancer patients . They performed several times during different stages of radiotherapy MRI scans. We need to be based on these MRI sacns Image , Propose a model using deep learning , Automatic segmentation of the inner intestine (stomach) And the stomach (intestines)

Commercial value ： In a medical scanning system , Locate healthy organs , Help improve cancer treatment

※ Match timeline

2022 4 month 14 Japan year - Start date . 2022 7 month 7 Japan year - Registration deadline . You must accept the competition rules before this date to participate in the competition . 2022 7 month 7 Japan year - Team merge deadline . This is the last day that participants can join or merge teams . 2022 7 month 14 Japan year - Final submission deadline .

※ Big bonus

The first prize ：12,000 dollar The second prize ：8,000 dollar The third prize ：5,000 dollar

3、 data EDA

Training data at a glance ：

Image visualization ：

4、baseline

Information provided this time baseline It is divided into hypre-parameter, data, model, Loss, train Five aspects .

A complete set of baseline The code is very clear , Very suitable for entry and subsequent code reuse .

5、 The evaluation index

50% Of Dice coefficient + 50% Of 3D Hausdorff distance.

6、Trick

Data to enhance ： Multi resolution training 、elastictransformer etc.

Model ： Advanced models ,Unet A series of variants of

Loss： Coefficient automatic learning strategy

Submit ： Multi model mixing, etc …