Monai version has been updated to 0.9. See what new functions it has

MONAI Update to 0.9 Version of the , How much do you use ？ Let's take a look at the important updates this time .

• MONAI Bundle：MONAI Bundles
• Object detection in medical images： Object detection in medical images
• Swin Transformers for 3D medical image analysis： be used for 3D Medical image analysis Swin Transformers
• New interactive segmentation components： New interactive subdivision components
• MetaTensor API preview：MetaTensor API preview

MONAI Bundle

MONAI Bundle The format defines a deep learning model of portable description （ file ）. The bundle includes the key information necessary in the model development lifecycle , And allow users and programs to understand the purpose and purpose of the model . Bundle and monai.bundle API The main advantage of ：

• Standardized packaging format for storing and sharing models,
  A standardized packaging format for storing and sharing models
• Structured configuration files for fast prototyping of deep learning workflows,
  Structured configuration file for deep learning workflow rapid prototyping ,
• Easy to program APIs to separate deep learning hyperparameter settings from the Python code,
  Easy to program API To set the depth learning superparameter with Python Code separation ,
• Flexible config components to allow for different low-level Python implementations,
  Flexible configuration of components to allow different low-level Python Realization ,
• Help to decouple the component details from higher level learning paradigms such as federated learning and AutoML.
  Help integrate component details with higher-level learning paradigms （ For example, federal learning and AutoML） decoupling 、

This part is suitable for technical leaders who have high requirements for code writing , I don't understand this kind of Xiaobai . More detailed tutorials MONAI bundle

Object detection in medical images

This release includes the basic components of the object localization and classification workflow . The initial development included 2D and 3D Bounding box processing 、RetinaNet Network block and architecture , And coordinate based preprocessing 、 Hard negative sampler and other commonly used practical modules .

Those who do object checking can pay attention to this part of the update .

Swin Transformers for 3D medical image analysis

Added Swin UNETR Model in MONAI To realize . This tutorial shows an example of multi organ segmentation using this state-of-the-art model , Among them, the weight comes from 5050 Time CT Scanned Swin UNETR Encoder （3D Swin Transformer） Self supervision pre training .

Implementation details in MOANI swin UNETR

I've been paying attention recently transformer Split children's shoes can try to run the code .

New interactive segmentation components

DeepEdit and NuClick The new components in the interactive segmentation workflow of equal depth learning have been integrated into the core code base . They are MONAILabel Basic building blocks for the latest functionality in .

Useful MONAI label Understand the marked data , I haven't understood how to use

MetaTensor API preview

Metadata related to major imaging patterns is important in many biomedical applications , Especially for MONAI The data-driven method that has been concerned . From this version , We are right. MONAI The data representation in has been significantly reconstructed . First step , Core data structure MetaTensor and MetaObj Implemented as function Preview . Further development of the functional branches will be provided in future milestone releases .

Interested in this update , It is recommended to upgrade and play ,PS： If you publish a paper with a previous version , Still under repair , It is suggested to build a new conda Environmental Science , Avoiding environmental changes cannot reproduce the experiment .

Articles are constantly updated , You can pay attention to the official account of WeChat 【 Medical image AI combat camp 】 Get the latest , The official account of the frontier technology in the field of medical image processing . Stick to the practice , Take you hand in hand to do the project , Play the game , Write a paper . All original articles provide theoretical explanation , Experimental code , experimental data . Only practice can grow faster , Pay attention to our , Learn together ~

I am a Tina, I'll see you on our next blog ~

Working during the day and writing at night , cough

If you think it's well written, finally , Please thumb up , Comment on , Collection . Or three times with one click