AI Toolkit for Healthcare Imaging

Project MONAI

Last update: Jan 7, 2023

Related tags

Deep Learning deep-learning python3 pytorch medical-image-computing medical-image-processing healthcare-imaging monai

Overview

Medical Open Network for AI

MONAI is a PyTorch-based, open-source framework for deep learning in healthcare imaging, part of PyTorch Ecosystem. Its ambitions are:

developing a community of academic, industrial and clinical researchers collaborating on a common foundation;
creating state-of-the-art, end-to-end training workflows for healthcare imaging;
providing researchers with the optimized and standardized way to create and evaluate deep learning models.

Features

The codebase is currently under active development. Please see the technical highlights and What's New of the current milestone release.

flexible pre-processing for multi-dimensional medical imaging data;
compositional & portable APIs for ease of integration in existing workflows;
domain-specific implementations for networks, losses, evaluation metrics and more;
customizable design for varying user expertise;
multi-GPU data parallelism support.

Installation

To install the current release, you can simply run:

pip install monai

For other installation methods (using the default GitHub branch, using Docker, etc.), please refer to the installation guide.

Getting Started

MedNIST demo and MONAI for PyTorch Users are available on Colab.

Examples and notebook tutorials are located at Project-MONAI/tutorials.

Technical documentation is available at docs.monai.io.

Contributing

For guidance on making a contribution to MONAI, see the contributing guidelines.

Community

Join the conversation on Twitter @ProjectMONAI or join our Slack channel.

Ask and answer questions over on MONAI's GitHub Discussions tab.

Links

Website: https://monai.io/
API documentation: https://docs.monai.io
Code: https://github.com/Project-MONAI/MONAI
Project tracker: https://github.com/Project-MONAI/MONAI/projects
Issue tracker: https://github.com/Project-MONAI/MONAI/issues
Wiki: https://github.com/Project-MONAI/MONAI/wiki
Test status: https://github.com/Project-MONAI/MONAI/actions
PyPI package: https://pypi.org/project/monai/
Weekly previews: https://pypi.org/project/monai-weekly/
Docker Hub: https://hub.docker.com/r/projectmonai/monai

Comments

add swin_unetr model
Signed-off-by: ahatamizadeh [email protected]

Fixes #3520 .

Description

This PR adds the Swin UNETR [1] model to MONAI.

[1]: Hatamizadeh, Ali, Vishwesh Nath, Yucheng Tang, Dong Yang, Holger Roth, and Daguang Xu. "Swin UNETR: Swin Transformers for Semantic Segmentation of Brain Tumors in MRI Images." arXiv preprint arXiv:2201.01266 (2022).

Swin UNETR is also used in self-supervised learning for 3D medical image segmentation [2].

[2]: Tang, Yucheng, Dong Yang, Wenqi Li, Holger Roth, Bennett Landman, Daguang Xu, Vishwesh Nath, and Ali Hatamizadeh. "Self-supervised pre-training of swin transformers for 3d medical image analysis." arXiv preprint arXiv:2111.14791 (2021).

Status

Ready

Types of changes

[x] Non-breaking change (fix or new feature that would not break existing functionality).

[x] New tests added to cover the changes.

[x] Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.

[x] Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.

[x] In-line docstrings updated.

[x] Documentation updated, tested make html command in the docs/ folder.

Feature request
opened by ahatamiz 68
3482 Add `ConfigComponent` for config parsing
Task step 2 of #3482 .

Description

This PR implemented the ConfigComponent feature for config parsing, it's for task 2 of #3482 . The whole proposal is in draft PR: https://github.com/Project-MONAI/MONAI/pull/3593.

Status

Ready

Types of changes

[x] Non-breaking change (fix or new feature that would not break existing functionality).

[ ] Breaking change (fix or new feature that would cause existing functionality to change).

[ ] New tests added to cover the changes.

[ ] Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.

[ ] Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.

[ ] In-line docstrings updated.

[ ] Documentation updated, tested make html command in the docs/ folder.
opened by Nic-Ma 59
Video dataset
Addresses: https://github.com/Project-MONAI/MONAI/issues/4746.

Description

Adds video datasets. Videos can be from file or capture device (e.g., webcam).

Requires a corresponding tutorial.

Status

Hold

Types of changes

[x] Non-breaking change (fix or new feature that would not break existing functionality).

[x] New tests added to cover the changes.
opened by rijobro 44
WIP: Fix failing A100 tests
Signed-off-by: Mohammad Adil [email protected]

Fixes #2357 .

Description

Change tolerance in tests to fix tests failing on A100 GPU.

Status

Work in progress

Types of changes

[x] Non-breaking change (fix or new feature that would not break existing functionality).

[ ] Breaking change (fix or new feature that would cause existing functionality to change).

[ ] New tests added to cover the changes.

[ ] Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.

[ ] Quick tests passed locally by running ./runtests.sh --quick --unittests.

[ ] In-line docstrings updated.

[ ] Documentation updated, tested make html command in the docs/ folder.

CI/CD
opened by madil90 41
5308 Add runtime cache support to `CacheDataset`
Fixes #5308 .

Description

This PR added runtime cache support in the CacheDataset.

Types of changes

[x] Non-breaking change (fix or new feature that would not break existing functionality).

[ ] Breaking change (fix or new feature that would cause existing functionality to change).

[ ] New tests added to cover the changes.

[ ] Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.

[ ] Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.

[ ] In-line docstrings updated.

[ ] Documentation updated, tested make html command in the docs/ folder.
opened by Nic-Ma 40
Develop MVP of model bundle
Is your feature request related to a problem? Please describe. Thanks for the interesting technical discussion with @ericspod @wyli @atbenmurray @rijobro , as we still have many unclear requirements and unknown use cases, we plan to develop the model package feature step by step, May adjust the design based on feedback during the development.

For the initial step, the core team aligned to develop a small but typical example for inference first, it will use JSON config files to define environments, components and workflow, save the config and model into TorchScript model. then other projects can easily reconstruct the exact same python program and parameters to reproduce the inference. When the small MVP is ready, will share and discuss within the team for the next steps.

I will try to implement the MVP referring to some existing solutions, like NVIDIA Clara MMAR, ignite online package, etc. Basic task steps:

[x] Include metadata (env / sys info, changelog, version, input / output data format, etc), configs, model weights, etc. in a model package example for review. PR: https://github.com/Project-MONAI/tutorials/pull/487

[x] Search specified python packages and build python instances from dictionary config with name / path & args. PR: https://github.com/Project-MONAI/MONAI/pull/3720

[x] Recursively parse configs in a dictionary with dependencies and executable code, for example: {"dataset": {"<name>": "Dataset", "<args>": {"data": "$load_datalist()"}}, "dataloader": {"<name>": "DataLoader", "<args>": {"data": "@dataset"}}}. PR: https://github.com/Project-MONAI/MONAI/pull/3818, https://github.com/Project-MONAI/MONAI/pull/3822

[x] Add support to save the raw config dictionaries into TorchScript model. PR: Project-MONAI/MONAI#3138 .

[x] Add schema mechanism to verify the content of folder structure, metadata.json content, etc. Refer to: https://json-schema.org/, https://github.com/Julian/jsonschema. PR: https://github.com/Project-MONAI/MONAI/pull/3865

[x] Add support to verify network input / output with fake data (data input is from metadata).

[x] Add mechanism to easily load JSON config files with override (can add YAML support in future), similar to example: https://hydra.cc/docs/tutorials/basic/your_first_app/config_groups/. PR: https://github.com/Project-MONAI/MONAI/pull/3832

[x] Add support to refer to other config item in the same config file or other confg files, referring to Hydra ideas: https://hydra.cc/docs/advanced/overriding_packages/.

[x] Complete the inference example MMAR for spleen segmentation task. PR: https://github.com/Project-MONAI/tutorials/pull/604

[x] Write user manual and detailed documentation. PR: https://github.com/Project-MONAI/MONAI/pull/3834 https://github.com/Project-MONAI/MONAI/pull/3982

[x] [*Optional] Upload the spleen MMAR example to huggingface(https://github.com/Project-MONAI/MONAI/discussions/3451).

[x] [*Optional] Support relative config level in the reference ID, for example: "test": "@###data#1", 1 # means current level, 2 ## means upper level, etc. PR: https://github.com/Project-MONAI/MONAI/pull/3974

[ ] [*Optional] Support customized ConfigItem and ReferenceResolver in the ConfigParser. PR: https://github.com/Project-MONAI/MONAI/pull/3980/

[x] _requires_ keyword for config component (https://github.com/Project-MONAI/MONAI/issues/3942).

[x] Import statement in bundle config (https://github.com/Project-MONAI/MONAI/issues/3966).

[x] Config python logging properties in a file. PR: https://github.com/Project-MONAI/MONAI/pull/3994

[ ] [*Optional] Specify rank ID for component to run only on some ranks, for example, saving checkpoint in rank 0.

Feature request
opened by Nic-Ma 38
Training instability with Dice Loss/Tversky Loss
I am training a 2D UNet to segment fetal MR images using MONAI and I have been observing some instability in the training when using MONAI Dice loss formulation. After some iteration, the loss jumps up and the network stops learning, as the gradients drop to zero. Here is an example (orange is loss on training set computed over 2D slices, blue is loss on validation computed over 3D volume).

After investigating several aspects (using the same deterministic seed), I've narrowed down the issue to the presence of the smooth term in both the numerator and denominator of the Dice Loss: f = 1.0 - (2.0 * intersection + smooth) / (denominator + smooth)

When using the formulation: f = 1.0 - (2.0 * intersection) / (denominator + smooth) without the smooth term in the numerator, the training was stable and no longer showed unexpected behaviour: [Note: this experiment was trained for much longer to make sure the jump would not appear later in the training]

The same pattern was observed also for the Tversky Loss, so it could be worth investigating the stability of the losses to identify the best default option.

Software version MONAI version: 0.1.0+84.ga683c4e.dirty Python version: 3.7.4 (default, Jul 9 2019, 03:52:42) [GCC 5.4.0 20160609] Numpy version: 1.18.2 Pytorch version: 1.4.0 Ignite version: 0.3.0

Training information Using MONAI PersistentCache 2D UNet (as default in MONAI) Adam optimiser, LR = 1e-3, no LR decay Batch size: 10

Other tests The following aspects were investigated but did not solve the instability issue:

Gradient clipping

Different optimisers (SGD, SGD + Momentum)

Transforming the binary segmentations to a two-channel approach ([background segmentation, foreground segmentation])

Choosing smooth = 1.0 as default here (https://github.com/MIC-DKFZ/nnUNet/blob/master/nnunet/training/loss_functions/dice_loss.py). However, this made the behaviour even more severe and the jump would happen sooner in the training.

The following losses were also investigated

Binary Cross Entropy --> stable

Dice Loss + Binary Cross Entropy --> unstable

Dice Loss (no smooth at numerator) + Binary Cross Entropy --> stable

Tversky Loss --> Unstable

Tversky Loss (no smooth at numerator) --> stable

bug
opened by martaranzini 38
Auto3D Task Module 1 (DataAnalyzer) for #4743
Description

Implemented a DataAnalyzer class to encapsulate data analysis module. As the beginning part of the auto3d/automl pipeline, the module shall find data and label from user inputs and generate a summary (dictionary) of data stats. The summary includes - file names, list, number of files; - dataset summary (basic information, image dimensions, number of classes, etc.); - individual data information (spacing, image size, number and size of the regions, etc.). The summary can be exported as a YAML file and a dictionary variable for use in Python

Example Usage:

from monai.apps.auto3d.data_analyzer import DataAnalyzer datalist = { "testing": [{"image": "image_003.nii.gz"}], "training": [ {"fold": 0, "image": "image_001.nii.gz", "label": "label_001.nii.gz"}, {"fold": 0, "image": "image_002.nii.gz", "label": "label_002.nii.gz"}, {"fold": 1, "image": "image_001.nii.gz", "label": "label_001.nii.gz"}, {"fold": 1, "image": "image_004.nii.gz", "label": "label_004.nii.gz"}, ], } dataroot = '/datasets' # the directory where you have the image files (in this example we're using nii.gz) analyser = DataAnalyzer(datalist, dataroot) datastat = analyser.get_all_case_stats() # it will also generate a data_stats.yaml that saves the stats

Status

Ready for Review Reference issue https://github.com/Project-MONAI/MONAI/issues/4743.

Types of changes

[x] Non-breaking change (fix or new feature that would not break existing functionality).

[x] New tests added to cover the changes.

[x] Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.

[x] In-line docstrings updated.

[x] Documentation updated, tested make html command in the docs/ folder.
opened by mingxin-zheng 37
3482 add `run` API for common training, evaluation and inference
Task step 7 of https://github.com/Project-MONAI/MONAI/issues/3482 .

Description

This PR added the run API to execute most common cases of supervised training, evaluation or inference from the config directly.

Status

Ready

Types of changes

[x] Non-breaking change (fix or new feature that would not break existing functionality).

[ ] Breaking change (fix or new feature that would cause existing functionality to change).

[ ] New tests added to cover the changes.

[ ] Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.

[ ] Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.

[ ] In-line docstrings updated.

[ ] Documentation updated, tested make html command in the docs/ folder.
opened by Nic-Ma 35
3482 3829 Add `ConfigParser` to recursively parse config content
Task step 3-2 of https://github.com/Project-MONAI/MONAI/issues/3482 . Fixes https://github.com/Project-MONAI/MONAI/issues/3829

Description

This PR added the ConfigParser to recursively parse config content.

Status

Ready

Types of changes

[x] Non-breaking change (fix or new feature that would not break existing functionality).

[ ] Breaking change (fix or new feature that would cause existing functionality to change).

[ ] New tests added to cover the changes.

[ ] Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.

[ ] Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.

[ ] In-line docstrings updated.

[ ] Documentation updated, tested make html command in the docs/ folder.
opened by Nic-Ma 35
Segmentation example on multiple labels

Is your feature request related to a problem? Please describe. The spleen segmentation example is incredibly helpful. Several of my projects involve multiple labels - e.g. the right and left lateral ventricle.

Describe the solution you'd like I'd greatly appreciate an example or guidance on how to train a model with multiple labels.

Additional context I'm pretty new to this, and suspect it's a pretty easy modification to the example.
question

opened by radiplab 33
SaveImaged not working after upgrading MONAI from 0.8.1 to 1.0.1
Describe the bug I upgraded the MONAI from 0.8.1 to 1.0.1. Then my original yaml config as following that works on 0.8.1 does not work anymore on 1.0.1.

_target_: SaveImaged, keys: model_output_key, meta_keys: [ 'labels_meta_dict' ], data_root_dir: label_root_dir, output_dir: pred_output_dir, output_postfix: 'pred', separate_folder: False

Any help would be greatly appreciated!

Expected behavior I expect the code to save the output to nii file.

Screenshots Error trackback followed in the end.

Environment

Ensuring you use the relevant python executable, please paste the output of:

python -c 'import monai; monai.config.print_debug_info()'

================================ Printing MONAI config...

MONAI version: 1.0.1 Numpy version: 1.24.1 Pytorch version: 1.13.1+cu117 MONAI flags: HAS_EXT = False, USE_COMPILED = False, USE_META_DICT = False MONAI rev id: 8271a193229fe4437026185e218d5b06f7c8ce69 MONAI file: /home/ubuntu/.cache/pypoetry/virtualenvs/qbio-qbdl-D8V9KvtL-py3.8/lib/python3.8/site-packages/monai/init.py

Optional dependencies: Pytorch Ignite version: NOT INSTALLED or UNKNOWN VERSION. Nibabel version: 3.2.2 scikit-image version: 0.19.3 Pillow version: 9.4.0 Tensorboard version: NOT INSTALLED or UNKNOWN VERSION. gdown version: NOT INSTALLED or UNKNOWN VERSION. TorchVision version: 0.14.1+cu117 tqdm version: 4.64.1 lmdb version: NOT INSTALLED or UNKNOWN VERSION. psutil version: NOT INSTALLED or UNKNOWN VERSION. pandas version: NOT INSTALLED or UNKNOWN VERSION. einops version: 0.5.0 transformers version: NOT INSTALLED or UNKNOWN VERSION. mlflow version: NOT INSTALLED or UNKNOWN VERSION. pynrrd version: NOT INSTALLED or UNKNOWN VERSION.

For details about installing the optional dependencies, please visit: https://docs.monai.io/en/latest/installation.html#installing-the-recommended-dependencies

================================ Printing system config...

psutil required for print_system_info

================================ Printing GPU config...

Num GPUs: 4 Has CUDA: True CUDA version: 11.7 cuDNN enabled: True cuDNN version: 8500 Current device: 0 Library compiled for CUDA architectures: ['sm_37', 'sm_50', 'sm_60', 'sm_70', 'sm_75', 'sm_80', 'sm_86'] GPU 0 Name: NVIDIA A10G GPU 0 Is integrated: False GPU 0 Is multi GPU board: False GPU 0 Multi processor count: 80 GPU 0 Total memory (GB): 22.2 GPU 0 CUDA capability (maj.min): 8.6 GPU 1 Name: NVIDIA A10G GPU 1 Is integrated: False GPU 1 Is multi GPU board: False GPU 1 Multi processor count: 80 GPU 1 Total memory (GB): 22.2 GPU 1 CUDA capability (maj.min): 8.6 GPU 2 Name: NVIDIA A10G GPU 2 Is integrated: False GPU 2 Is multi GPU board: False GPU 2 Multi processor count: 80 GPU 2 Total memory (GB): 22.2 GPU 2 CUDA capability (maj.min): 8.6 GPU 3 Name: NVIDIA A10G GPU 3 Is integrated: False GPU 3 Is multi GPU board: False GPU 3 Multi processor count: 80 GPU 3 Total memory (GB): 22.2 GPU 3 CUDA capability (maj.min): 8.6

Additional context Add any other context about the problem here. Error traceback:

Traceback (most recent call last): File "/home/ubuntu/.cache/pypoetry/virtualenvs/qbio-qbdl-D8V9KvtL-py3.8/lib/python3.8/site-packages/monai/transforms/transform.py", line 91, in apply_transform return _apply_transform(transform, data, unpack_items) File "/home/ubuntu/.cache/pypoetry/virtualenvs/qbio-qbdl-D8V9KvtL-py3.8/lib/python3.8/site-packages/monai/transforms/transform.py", line 55, in _apply_transform return transform(parameters) File "/home/ubuntu/.cache/pypoetry/virtualenvs/qbio-qbdl-D8V9KvtL-py3.8/lib/python3.8/site-packages/monai/transforms/io/dictionary.py", line 293, in call self.saver(img=d[key], meta_data=meta_data) File "/home/ubuntu/.cache/pypoetry/virtualenvs/qbio-qbdl-D8V9KvtL-py3.8/lib/python3.8/site-packages/monai/transforms/io/array.py", line 423, in call subject = meta_data[Key.FILENAME_OR_OBJ] if meta_data else str(self._data_index) KeyError: 'filename_or_obj'
opened by moonforsun 0
[Feature Request]: CLIP Driven Universal Model
Contrastive Language-Image Pre-training (CLIP) Driven Models and Partially Supervised Learning for Medical Image Segmentation

This issue is to discuss adding the CLIP-Driven Universal Model Features to MONAI.

This will provide the implementation of top 1 solution in Medical Segmentation Decathlon (MSD).

Potential assignee: @tangy5

CLIP-Driven Universal Model

Universal Model is the first framework for both organ segmentation and tumor detection. The method achieves and provide solution to the top spot of MSD competition leaderboard.

Paper

CLIP-Driven Universal Model for Organ Segmentation and Tumor Detection
Pre-print: https://arxiv.org/pdf/2301.00785.pdf

Key features

The implementation will bring several new feature as follows:

Universal Model: one model to detect and segment all abdominal organs and all types of tumors (Liver tumor, kidney tumor, Lung nodule, Pancreas tumor, hepatic vessel tumor, colon tumor).

Language model (CLIP) and text-driven embeddings boost medical image analysis.

Training Partial labelled datasets.

Incremental learning: Users can continue to train new segmentation classes using the current trained model without catastrophic forgetting.

⏳ Dataset: The Universal Model is trained with following datasets

01 Multi-Atlas Labeling Beyond the Cranial Vault - Workshop and Challenge (BTCV)

02 Pancreas-CT TCIA

03 Combined Healthy Abdominal Organ Segmentation (CHAOS)

04 Liver Tumor Segmentation Challenge (LiTS)

05 Kidney and Kidney Tumor Segmentation (KiTS)

06 Liver segmentation (3D-IRCADb)

07 WORD: A large scale dataset, benchmark and clinical applicable study for abdominal organ segmentation from CT image

08 AbdomenCT-1K

09 Multi-Modality Abdominal Multi-Organ Segmentation Challenge (AMOS)

10-15 Decathlon (Liver, Lung, Pancreas, HepaticVessel, Spleen, Colon)

16 CT volumes with multiple organ segmentations (CT-ORG)

17 13 AbdomenCT 12organ

Implementation plans

[ ] Transformations (pre-processing) for partial labelled datasets: “PartialLabelTransfer”, etc

[ ] Segmentation backbone with CLIP embedding, text-driven segmentor: plug-and-play CLIP embedding and text encoder.

[ ] Tutorial for training and inference of Universal Model.

[ ] Tutorial for demonstrating partial supervised learning and incremental learning.

[ ] Model release: Bundle for Model Zoo for publishing the trained universal model to segment all types of tumours and abdominal organs.

More Details of the Feature Methodology:

Universal Model:

CLIP Driven and text-driven segmentor:

Partial Supervised Learning:

Incremental Leraning:

Detailed steps of implantation will provide after open discussion.

Welcome all suggestions and comments!

@ljwztc @MrGiovanni
Design discussions Contribution wanted Feature request
opened by tangy5 0
invalid pointer error
Hello I install monai via

pip install monai-weekly

on Ubuntu 18 python 3.9 torch-1.12.0+cu116 When i do

import monai

I get error

free(): invalid pointer Aborted (core dumped)
opened by jakubMitura14 0
[pre-commit.ci] pre-commit suggestions
updates:

github.com/pre-commit/pre-commit-hooks: v4.3.0 → v4.4.0

github.com/asottile/pyupgrade: v2.38.2 → v3.3.1

github.com/hadialqattan/pycln: v2.1.1 → v2.1.2
opened by pre-commit-ci[bot] 0
'dimensions' -> 'spatial_dims'
VarAutoEncoder.init() got an unexpected keyword argument 'dimensions' because now it is called spatial_dims.

Fixes # .

Description

A few sentences describing the changes proposed in this pull request.

Types of changes

[x] Non-breaking change (fix or new feature that would not break existing functionality).

[ ] Breaking change (fix or new feature that would cause existing functionality to change).

[ ] New tests added to cover the changes.

[ ] Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.

[ ] Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.

[ ] In-line docstrings updated.

[ ] Documentation updated, tested make html command in the docs/ folder.
opened by ShadowTwin41 8
5775 Fix `_get_latest_bundle_version` issue on Windows
Signed-off-by: Yiheng Wang [email protected]

Fixes #5775 .

Description

This PR fixes the issue of _get_latest_bundle_version on Windows (os.path.join will produce backslash which will create a wrong url) Thanks @SachidanandAlle for finding this issue.

Types of changes

[x] Non-breaking change (fix or new feature that would not break existing functionality).

[ ] Breaking change (fix or new feature that would cause existing functionality to change).

[ ] New tests added to cover the changes.

[ ] Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.

[ ] Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.

[ ] In-line docstrings updated.

[ ] Documentation updated, tested make html command in the docs/ folder.
opened by yiheng-wang-nv 1

Releases(1.1.0)

1.1.0(Dec 19, 2022)
Added

Hover-Net based digital pathology workflows including new network, loss, postprocessing, metric, training, and inference modules

Various enhancements for Auto3dSeg AutoRunner including template caching, selection, and a dry-run mode nni_dry_run

Various enhancements for Auto3dSeg algo templates including new state-of-the-art configurations, optimized GPU memory utilization

New bundle API and configurations to support experiment management including MLFlowHandler

New bundle.script API to support model zoo query and download

LossMetric metric to compute loss as cumulative metric measurement

Transforms and base transform APIs including RandomizableTrait and MedianSmooth

runtime_cache option for CacheDataset and the derived classes to allow for shared caching on the fly

Flexible name formatter for SaveImage transform

pending_operations MetaTensor property and basic APIs for lazy image resampling

Contrastive sensitivity for SSIM metric

Extensible backbones for FlexibleUNet

Generalize SobelGradients to 3D and any spatial axes

warmup_multiplier option for WarmupCosineSchedule

F beta score metric based on confusion matrix metric

Support of key overwriting in LambdaD

Basic premerge tests for Python 3.11

Unit and integration tests for CUDA 11.6, 11.7 and A100 GPU

DataAnalyzer handles minor image-label shape inconsistencies

Fixed

Review and enhance previously untyped APIs with additional type annotations and casts

switch_endianness in LoadImage now supports tensor input

Reduced memory footprint for various Auto3dSeg tests

Issue of @ in monai.bundle.ReferenceResolver

Compatibility issue with ITK-Python 5.3 (converting itkMatrixF44 for default collate)

Inconsistent of sform and qform when using different backends for SaveImage

MetaTensor.shape call now returns a torch.Size instead of tuple

Issue of channel reduction in GeneralizedDiceLoss

Issue of background handling before softmax in DiceFocalLoss

Numerical issue of LocalNormalizedCrossCorrelationLoss

Issue of incompatible view size in ConfusionMatrixMetric

NetAdapter compatibility with Torchscript

Issue of extract_levels in RegUNet

Optional bias_downsample in ResNet

dtype overflow for ShiftIntensity transform

Randomized transforms such as RandCuCIM now inherit RandomizableTrait

fg_indices.size compatibility issue in generate_pos_neg_label_crop_centers

Issue when inverting ToTensor

Issue of capital letters in filename suffixes check in LoadImage

Minor tensor compatibility issues in apps.nuclick.transforms

Issue of float16 in verify_net_in_out

std variable type issue for RandRicianNoise

DataAnalyzer accepts None as label key and checks empty labels

iter_patch_position now has a smaller memory footprint

CumulativeAverage has been refactored and enhanced to allow for simple tracking of metric running stats.

Multi-threading issue for MLFlowHandler

Changed

Printing a MetaTensor now generates a less verbose representation

DistributedSampler raises a ValueError if there are too few devices

OpenCV and VideoDataset modules are loaded lazily to avoid dependency issues

device in monai.engines.Workflow supports string values

Activations and AsDiscrete take kwargs as additional arguments

DataAnalyzer is now more efficient and writes summary stats before detailed all case stats

Base Docker image upgraded to nvcr.io/nvidia/pytorch:22.10-py3 from nvcr.io/nvidia/pytorch:22.09-py3

Simplified Conda environment file environment-dev.yml

Versioneer dependency upgraded to 0.23 from 0.19

Deprecated

NibabelReader input argument dtype is deprecated, the reader will use the original dtype of the image

Removed

Support for PyTorch 1.7

Source code(tar.gz)
Source code(zip)
1.0.1(Oct 24, 2022)
Fixes

DiceCELoss for multichannel targets

Auto3DSeg DataAnalyzer out-of-memory error and other minor issues

An optional flag issue in the RetinaNet detector

An issue with output offset for Spacing

A LoadImage issue when track_meta is False

1D data output error in VarAutoEncoder

An issue with resolution computing in ImageStats

Added

Flexible min/max pixdim options for Spacing

Upsample mode deconvgroup and optional kernel sizes

Docstrings for gradient-based saliency maps

Occlusion sensitivity to use sliding window inference

Enhanced Gaussian window and device assignments for sliding window inference

Multi-GPU support for MonaiAlgo

ClientAlgoStats and MonaiAlgoStats for federated summary statistics

MetaTensor support for OneOf

Add a file check for bundle logging config

Additional content and an authentication token option for bundle info API

An anti-aliasing option for Resized

SlidingWindowInferer adaptive device based on cpu_thresh

SegResNetDS with deep supervision and non-isotropic kernel support

Premerge tests for Python 3.10

Changed

Base Docker image upgraded to nvcr.io/nvidia/pytorch:22.09-py3 from nvcr.io/nvidia/pytorch:22.08-py3

Replace None type metadata content with "none" for collate_fn compatibility

HoVerNet Mode and Branch to independent StrEnum

Automatically infer device from the first item in random elastic deformation dict

Add channel dim in ComputeHoVerMaps and ComputeHoVerMapsd

Remove batch dim in SobelGradients and SobelGradientsd

Deprecated

Deprecating compute_meandice, compute_meaniou in monai.metrics, in favor of compute_dice and compute_iou respectively

Source code(tar.gz)
Source code(zip)
1.0.0(Sep 16, 2022)
Added

monai.auto3dseg base APIs and monai.apps.auto3dseg components for automated machine learning (AutoML) workflow

monai.fl module with base APIs and MonaiAlgo for federated learning client workflow

An initial backwards compatibility guide

Initial release of accelerated MRI reconstruction components, including CoilSensitivityModel

Support of MetaTensor and new metadata attributes for various digital pathology components

Various monai.bundle enhancements for MONAI model-zoo usability, including config debug mode and get_all_bundles_list

new monai.transforms components including SignalContinuousWavelet for 1D signal, ComputeHoVerMaps for digital pathology, and SobelGradients for spatial gradients

VarianceMetric and LabelQualityScore metrics for active learning

Dataset API for real-time stream and videos

Several networks and building blocks including FlexibleUNet and HoVerNet

MeanIoUHandler and LogfileHandler workflow event handlers

WSIReader with the TiffFile backend

Multi-threading in WSIReader with cuCIM backend

get_stats API in monai.engines.Workflow

prune_meta_pattern in monai.transforms.LoadImage

max_interactions for deepedit interaction workflow

Various profiling utilities in monai.utils.profiling

Changed

Base Docker image upgraded to nvcr.io/nvidia/pytorch:22.08-py3 from nvcr.io/nvidia/pytorch:22.06-py3

Optionally depend on PyTorch-Ignite v0.4.10 instead of v0.4.9

The cache-based dataset now matches the transform information when read/write the cache

monai.losses.ContrastiveLoss now infers batch_size during forward()

Rearrange the spatial axes in RandSmoothDeform transforms following PyTorch's convention

Unified several environment flags into monai.utils.misc.MONAIEnvVars

Simplified __str__ implementation of MetaTensor instead of relying on the __repr__ implementation

Fixed

Improved error messages when both monai and monai-weekly are pip-installed

Inconsistent pseudo number sequences for different num_workers in DataLoader

Issue of repeated sequences for monai.data.ShuffleBuffer

Issue of not preserving the physical extent in monai.transforms.Spacing

Issue of using inception_v3 as the backbone of monai.networks.nets.TorchVisionFCModel

Index device issue for monai.transforms.Crop

Efficiency issue when converting the array dtype and contiguous memory

Deprecated

Addchannel and AsChannelFirst transforms in favor of EnsureChannelFirst

monai.apps.pathology.data components in favor of the corresponding components from monai.data

monai.apps.pathology.handlers in favor of the corresponding components from monai.handlers

Removed

Status section in the pull request template in favor of the pull request draft mode

monai.engines.BaseWorkflow

ndim and dimensions arguments in favor of spatial_dims

n_classes, num_classes arguments in AsDiscrete in favor of to_onehot

logit_thresh, threshold_values arguments in AsDiscrete in favor of threshold

torch.testing.assert_allclose in favor of tests.utils.assert_allclose

Source code(tar.gz)
Source code(zip)
0.9.1(Jul 25, 2022)
Added

Support of monai.data.MetaTensor as core data structure across the modules

Support of inverse in array-based transforms

monai.apps.TciaDataset APIs for The Cancer Imaging Archive (TCIA) datasets, including a pydicom-backend reader

Initial release of components for MRI reconstruction in monai.apps.reconstruction, including various FFT utilities

New metrics and losses, including mean IoU and structural similarity index

monai.utils.StrEnum class to simplify Enum-based type annotations

Changed

Base Docker image upgraded to nvcr.io/nvidia/pytorch:22.06-py3 from nvcr.io/nvidia/pytorch:22.04-py3

Optionally depend on PyTorch-Ignite v0.4.9 instead of v0.4.8

Fixed

Fixed issue of not skipping post activations in Convolution when input arguments are None

Fixed issue of ignoring dropout arguments in DynUNet

Fixed issue of hard-coded non-linear function in ViT classification head

Fixed issue of in-memory config overriding with monai.bundle.ConfigParser.update

2D SwinUNETR incompatible shapes

Fixed issue with monai.bundle.verify_metadata not raising exceptions

Fixed issue with monai.transforms.GridPatch returns inconsistent type location when padding

Wrong generalized Dice score metric when denominator is 0 but prediction is non-empty

Docker image build error due to NGC CLI upgrade

Optional default value when parsing id unavailable in a ConfigParser instance

Immutable data input for the patch-based WSI datasets

Deprecated

*_transforms and *_meta_dict fields in dictionary-based transforms in favor of MetaTensor

meta_keys, meta_key_postfix, src_affine arguments in various transforms, in favor of MetaTensor

AsChannelFirst and AddChannel, in favor of EnsureChannelFirst transform

Source code(tar.gz)
Source code(zip)
0.9.0(Jun 13, 2022)
Added

monai.bundle primary module with a ConfigParser and command-line interfaces for config-based workflows

Initial release of MONAI bundle specification

Initial release of volumetric image detection modules including bounding boxes handling, RetinaNet-based architectures

API preview monai.data.MetaTensor

Unified monai.data.image_writer to support flexible IO backends including an ITK writer

Various new network blocks and architectures including SwinUNETR

DeepEdit interactive training/validation workflow

NuClick interactive segmentation transforms

Patch-based readers and datasets for whole-slide imaging

New losses and metrics including SurfaceDiceMetric, GeneralizedDiceFocalLoss

New pre-processing transforms including RandIntensityRemap, SpatialResample

Multi-output and slice-based inference for SlidingWindowInferer

NrrdReader for NRRD file support

Torchscript utilities to save models with meta information

Gradient-based visualization module SmoothGrad

Automatic regular source code scanning for common vulnerabilities and coding errors

Changed

Simplified TestTimeAugmentation using de-collate and invertible transforms APIs

Refactoring monai.apps.pathology modules into monai.handlers and monai.transforms

Flexible activation and normalization layers for TopologySearch and DiNTS

Anisotropic first layers for 3D resnet

Flexible ordering of activation, normalization in UNet

Enhanced performance of connected-components analysis using Cupy

INSTANCE_NVFUSER for enhanced performance in 3D instance norm

Support of string representation of dtype in convert_data_type

Added new options iteration_log, iteration_log to the logging handlers

Base Docker image upgraded to nvcr.io/nvidia/pytorch:22.04-py3 from nvcr.io/nvidia/pytorch:21.10-py3

collate_fn generates more data-related debugging info with dev_collate

Fixed

Unified the spellings of "meta data", "metadata", "meta-data" to "metadata"

Various inaccurate error messages when input data are in invalid shapes

Issue of computing symmetric distances in compute_average_surface_distance

Unnecessary layer self.conv3 in UnetResBlock

Issue of torchscript compatibility for ViT and self-attention blocks

Issue of hidden layers in UNETR

allow_smaller in spatial cropping transforms

Antialiasing in Resize

Issue of bending energy loss value at different resolutions

kwargs_read_csv in CSVDataset

In-place modification in Metric reduction

wrap_array for ensure_tuple

Contribution guide for introducing new third-party dependencies

Removed

Deprecated nifti_writer, png_writer in favor of monai.data.image_writer

Support for PyTorch 1.6

Source code(tar.gz)
Source code(zip)
0.8.1(Feb 16, 2022)
Added

Support of spatial 2D for ViTAutoEnc

Support of dataframe object input in CSVDataset

Support of tensor backend for Orientation

Support of configurable delimiter for CSV writers

A base workflow API

DataFunc API for dataset-level preprocessing

write_scalar API for logging with additional engine parameter in TensorBoardHandler

Enhancements for NVTX Range transform logging

Enhancements for set_determinism

Performance enhancements in the cache-based datasets

Configurable metadata keys for monai.data.DatasetSummary

Flexible kwargs for WSIReader

Logging for the learning rate schedule handler

GridPatchDataset as subclass of monai.data.IterableDataset

is_onehot option in KeepLargestConnectedComponent

channel_dim in the image readers and support of stacking images with channels

Support of matshow3d with given channel_dim

Skipping workflow run if epoch length is 0

Enhanced CacheDataset to avoid duplicated cache items

save_state utility function

Changed

Optionally depend on PyTorch-Ignite v0.4.8 instead of v0.4.6

monai.apps.mmars.load_from_mmar defaults to the latest version

Fixed

Issue when caching large items with pickle

Issue of hard-coded activation functions in ResBlock

Issue of create_file_name assuming local disk file creation

Issue of WSIReader when the backend is TiffFile

Issue of deprecated_args when the function signature contains kwargs

Issue of channel_wise computations for the intensity-based transforms

Issue of inverting OneOf

Issue of removing temporary caching file for the persistent dataset

Error messages when reader backend is not available

Output type casting issue in ScaleIntensityRangePercentiles

Various docstring typos and broken URLs

mode in the evaluator engine

Ordering of Orientation and Spacing in monai.apps.deepgrow.dataset

Removed

Additional deep supervision modules in DynUnet

Deprecated reduction argument for ContrastiveLoss

Decollate warning in Workflow

Unique label exception in ROCAUCMetric

Logger configuration logic in the event handlers

Source code(tar.gz)
Source code(zip)
0.8.0(Nov 25, 2021)
Added

Overview of new features in v0.8

Network modules for differentiable neural network topology search (DiNTS)

Multiple Instance Learning transforms and models for digital pathology WSI analysis

Vision transformers for self-supervised representation learning

Contrastive loss for self-supervised learning

Finalized major improvements of 200+ components in monai.transforms to support input and backend in PyTorch and NumPy

Initial registration module benchmarking with GlobalMutualInformationLoss as an example

monai.transforms documentation with visual examples and the utility functions

Event handler for MLfLow integration

Enhanced data visualization functions including blend_images and matshow3d

RandGridDistortion and SmoothField in monai.transforms

Support of randomized shuffle buffer in iterable datasets

Performance review and enhancements for data type casting

Cumulative averaging API with distributed environment support

Module utility functions including require_pkg and pytorch_after

Various usability enhancements such as allow_smaller when sampling ROI and wrap_sequence when casting object types

tifffile support in WSIReader

Regression tests for the fast training workflows

Various tutorials and demos including educational contents at MONAI Bootcamp 2021

Changed

Base Docker image upgraded to nvcr.io/nvidia/pytorch:21.10-py3 from nvcr.io/nvidia/pytorch:21.08-py3

Decoupled TraceKeys and TraceableTransform APIs from InvertibleTransform

Skipping affine-based resampling when resample=False in NiftiSaver

Deprecated threshold_values: bool and num_classes: int in AsDiscrete

Enhanced apply_filter for spatially 1D, 2D and 3D inputs with non-separable kernels

Logging with logging in downloading and model archives in monai.apps

API documentation site now defaults to stable instead of latest

skip-magic-trailing-comma in coding style enforcements

Pre-merge CI pipelines now include unit tests with Nvidia Ampere architecture

Removed

Support for PyTorch 1.5

The deprecated DynUnetV1 and the related network blocks

GitHub self-hosted CI/CD pipelines for package releases

Fixed

Support of path-like objects as file path inputs in most modules

Issue of decollate_batch for dictionary of empty lists

Typos in documentation and code examples in various modules

Issue of no available keys when allow_missing_keys=True for the MapTransform

Issue of redundant computation when normalization factors are 0.0 and 1.0 in ScaleIntensity

Incorrect reports of registered readers in ImageReader

Wrong numbering of iterations in StatsHandler

Naming conflicts in network modules and aliases

Incorrect output shape when reduction="none" in FocalLoss

Various usability issues reported by users

Source code(tar.gz)
Source code(zip)
0.7.0(Sep 24, 2021)
Added

Overview of new features in v0.7

Initial phase of major usability improvements in monai.transforms to support input and backend in PyTorch and NumPy

Performance enhancements, with profiling and tuning guides for typical use cases

Reproducing training modules and workflows of state-of-the-art Kaggle competition solutions

24 new transforms, including

OneOf meta transform

DeepEdit guidance signal transforms for interactive segmentation

Transforms for self-supervised pre-training

Integration of NVIDIA Tools Extension (NVTX)

Integration of cuCIM

Stain normalization and contextual grid for digital pathology

Transchex network for vision-language transformers for chest X-ray analysis

DatasetSummary utility in monai.data

WarmupCosineSchedule

Deprecation warnings and documentation support for better backwards compatibility

Padding with additional kwargs and different backend API

Additional options such as dropout and norm in various networks and their submodules

Changed

Base Docker image upgraded to nvcr.io/nvidia/pytorch:21.08-py3 from nvcr.io/nvidia/pytorch:21.06-py3

Deprecated input argument n_classes, in favor of num_classes

Deprecated input argument dimensions and ndims, in favor of spatial_dims

Updated the Sphinx-based documentation theme for better readability

NdarrayTensor type is replaced by NdarrayOrTensor for simpler annotations

Self-attention-based network blocks now support both 2D and 3D inputs

Removed

The deprecated TransformInverter, in favor of monai.transforms.InvertD

GitHub self-hosted CI/CD pipelines for nightly and post-merge tests

monai.handlers.utils.evenly_divisible_all_gather

monai.handlers.utils.string_list_all_gather

Fixed

A Multi-thread cache writing issue in LMDBDataset

Output shape convention inconsistencies of the image readers

Output directory and file name flexibility issue for NiftiSaver, PNGSaver

Requirement of the label field in test-time augmentation

Input argument flexibility issues for ThreadDataLoader

Decoupled Dice and CrossEntropy intermediate results in DiceCELoss

Improved documentation, code examples, and warning messages in various modules

Various usability issues reported by users

Source code(tar.gz)
Source code(zip)
0.6.0(Jul 8, 2021)
Added

Overview document for feature highlights in v0.6

10 new transforms, a masked loss wrapper, and a NetAdapter for transfer learning

APIs to load networks and pre-trained weights from Clara Train Medical Model ARchives (MMARs)

Base metric and cumulative metric APIs, 4 new regression metrics

Initial CSV dataset support

Decollating mini-batch as the default first postprocessing step

Initial backward compatibility support via monai.utils.deprecated

Attention-based vision modules and UNETR for segmentation

Generic module loaders and Gaussian mixture models using the PyTorch JIT compilation

Inverse of image patch sampling transforms

Network block utilities get_[norm, act, dropout, pool]_layer

unpack_items mode for apply_transform and Compose

New event INNER_ITERATION_STARTED in the deepgrow interactive workflow

set_data API for cache-based datasets to dynamically update the dataset content

Fully compatible with PyTorch 1.9

--disttests and --min options for runtests.sh

Initial support of pre-merge tests with Nvidia Blossom system

Changed

Base Docker image upgraded to nvcr.io/nvidia/pytorch:21.06-py3 from nvcr.io/nvidia/pytorch:21.04-py3

Optionally depend on PyTorch-Ignite v0.4.5 instead of v0.4.4

Unified the demo, tutorial, testing data to the project shared drive, and Project-MONAI/MONAI-extra-test-data

Unified the terms: post_transform is renamed to postprocessing, pre_transform is renamed to preprocessing

Unified the postprocessing transforms and event handlers to accept the "channel-first" data format

evenly_divisible_all_gather and string_list_all_gather moved to monai.utils.dist

Removed

Support of 'batched' input for postprocessing transforms and event handlers

TorchVisionFullyConvModel

set_visible_devices utility function

SegmentationSaver and TransformsInverter handlers

Fixed

Issue of handling big-endian image headers

Multi-thread issue for non-random transforms in the cache-based datasets

Persistent dataset issue when multiple processes sharing a non-exist cache location

Typing issue with Numpy 1.21.0

Loading checkpoint with both model and optmizier using CheckpointLoader when strict_shape=False

SplitChannel has different behaviour depending on numpy/torch inputs

Transform pickling issue caused by the Lambda functions

Issue of filtering by name in generate_param_groups

Inconsistencies in the return value types of class_activation_maps

Various docstring typos

Various usability enhancements in monai.transforms

Source code(tar.gz)
Source code(zip)
0.5.3(Jun 1, 2021)
Changed

Project default branch renamed to dev from master

Base Docker image upgraded to nvcr.io/nvidia/pytorch:21.02-py3 from nvcr.io/nvidia/pytorch:21.04-py3

Enhanced type checks for the iteration_metric handler

Enhanced PersistentDataset to use tempfile during caching computation

Enhanced various info/error messages

Enhanced performance of RandAffine

Enhanced performance of SmartCacheDataset

Optionally requires cucim when the platform is Linux

Default device of TestTimeAugmentation changed to cpu

Fixed

Download utilities now provide better default parameters

Duplicated key_transforms in the patch-based transforms

A multi-GPU issue in ClassificationSaver

A default meta_data issue in SpacingD

Dataset caching issue with the persistent data loader workers

A memory issue in permutohedral_cuda

Dictionary key issue in CopyItemsd

box_start and box_end parameters for deepgrow SpatialCropForegroundd

Tissue mask array transpose issue in MaskedInferenceWSIDataset

Various type hint errors

Various docstring typos

Added

Support of to_tensor and device arguments for TransformInverter

Slicing options with SpatialCrop

Class name alias for the networks for backward compatibility

k_divisible option for CropForeground

map_items option for Compose

Warnings of inf and nan for surface distance computation

A print_log flag to the image savers

Basic testing pipelines for Python 3.9

Source code(tar.gz)
Source code(zip)
0.5.0(Apr 13, 2021)
Added

Overview document for feature highlights in v0.5.0

Invertible spatial transforms

InvertibleTransform base APIs

Batch inverse and decollating APIs

Inverse of Compose

Batch inverse event handling

Test-time augmentation as an application

Initial support of learning-based image registration:

Bending energy, LNCC, and global mutual information loss

Fully convolutional architectures

Dense displacement field, dense velocity field computation

Warping with high-order interpolation with C++/CUDA implementations

Deepgrow modules for interactive segmentation:

Workflows with simulations of clicks

Distance-based transforms for guidance signals

Digital pathology support:

Efficient whole slide imaging IO and sampling with Nvidia cuCIM and SmartCache

FROC measurements for lesion

Probabilistic post-processing for lesion detection

TorchVision classification model adaptor for fully convolutional analysis

12 new transforms, grid patch dataset, ThreadDataLoader, EfficientNets B0-B7

4 iteration events for the engine for finer control of workflows

New C++/CUDA extensions:

Conditional random field

Fast bilateral filtering using the permutohedral lattice

Metrics summary reporting and saving APIs

DiceCELoss, DiceFocalLoss, a multi-scale wrapper for segmentation loss computation

Data loading utilities：

decollate_batch

PadListDataCollate with inverse support

Support of slicing syntax for Dataset

Initial Torchscript support for the loss modules

Learning rate finder

Allow for missing keys in the dictionary-based transforms

Support of checkpoint loading for transfer learning

Various summary and plotting utilities for Jupyter notebooks

Contributor Covenant Code of Conduct

Major CI/CD enhancements covering the tutorial repository

Fully compatible with PyTorch 1.8

Initial nightly CI/CD pipelines using Nvidia Blossom Infrastructure

Changed

Enhanced list_data_collate error handling

Unified iteration metric APIs

densenet* extensions are renamed to DenseNet*

se_res* network extensions are renamed to SERes*

Transform base APIs are rearranged into compose, inverse, and transform

_do_transform flag for the random augmentations is unified via RandomizableTransform

Decoupled post-processing steps, e.g. softmax, to_onehot_y, from the metrics computations

Moved the distributed samplers to monai.data.samplers from monai.data.utils

Engine's data loaders now accept generic iterables as input

Workflows now accept additional custom events and state properties

Various type hints according to Numpy 1.20

Refactored testing utility runtests.sh to have --unittest and --net integration tests options

Base Docker image upgraded to nvcr.io/nvidia/pytorch:21.02-py3 from nvcr.io/nvidia/pytorch:20.10-py3

Docker images are now built with self-hosted environments

Primary contact email updated to [email protected]

Now using GitHub Discussions as the primary communication forum

Removed

Compatibility tests for PyTorch 1.5.x

Format specific loaders, e.g. LoadNifti, NiftiDataset

Assert statements from non-test files

from module import * statements, addressed flake8 F403

Fixed

Uses American English spelling for code, as per PyTorch

Code coverage now takes multiprocessing runs into account

SmartCache with initial shuffling

ConvertToMultiChannelBasedOnBratsClasses now supports channel-first inputs

Checkpoint handler to save with non-root permissions

Fixed an issue for exiting the distributed unit tests

Unified DynUNet to have single tensor output w/o deep supervision

SegmentationSaver now supports user-specified data types and a squeeze_end_dims flag

Fixed *Saver event handlers output filenames with a data_root_dir option

Load image functions now ensure little-endian

Fixed the test runner to support regex-based test case matching

Usability issues in the event handlers

Source code(tar.gz)
Source code(zip)
0.4.0(Dec 15, 2020)
Added

Overview document for feature highlights in v0.4.0

Torchscript support for the net modules

New networks and layers:

Discrete Gaussian kernels

Hilbert transform and envelope detection

Swish and Mish activation

Acti-norm-dropout block

Upsampling layer

Autoencoder, Variational autoencoder

FCNet

Support of initialisation from pre-trained weights for densenet, SENet, multichannel AHNet

Layer-wise learning rate API

New model metrics and event handlers based on occlusion sensitivity, confusion matrix, surface distance

CAM/GradCAM/GradCAM++

File format-agnostic image loader APIs with Nibabel, ITK readers

Enhancements for dataset partition, cross-validation APIs

New data APIs:

LMDB-based caching dataset

Cache-N-transforms dataset

Iterable dataset

Patch dataset

Weekly PyPI release

Fully compatible with PyTorch 1.7

CI/CD enhancements:

Skip, speed up, fail fast, timed, quick tests

Distributed training tests

Performance profiling utilities

New tutorials and demos:

Autoencoder, VAE tutorial

Cross-validation demo

Model interpretability tutorial

COVID-19 Lung CT segmentation challenge open-source baseline

Threadbuffer demo

Dataset partitioning tutorial

Layer-wise learning rate demo

MONAI Bootcamp 2020

Changed

Base Docker image upgraded to nvcr.io/nvidia/pytorch:20.10-py3 from nvcr.io/nvidia/pytorch:20.08-py3

Backwards Incompatible Changes

monai.apps.CVDecathlonDataset is extended to a generic monai.apps.CrossValidation with an dataset_cls option

Cache dataset now requires a monai.transforms.Compose instance as the transform argument

Model checkpoint file name extensions changed from .pth to .pt

Readers' get_spatial_shape returns a numpy array instead of list

Decoupled postprocessing steps such as sigmoid, to_onehot_y, mutually_exclusive, logit_thresh from metrics and event handlers, the postprocessing steps should be used accordingly before computing the metrics

ConfusionMatrixMetric and DiceMetric computation now returns an additional not_nans flag to indicate valid results

UpSample optional mode supports "deconv", "nontrainable", "pixelshuffle"; interp_mode is only used when mode is "nontrainable"

SegResNet optional upsample_mode now supports "deconv", "nontrainable", "pixelshuffle"

monai.transforms.Compose class inherits monai.transforms.Transform

In Rotate, Rotated, RandRotate, RandRotated transforms, the angle related parameters are interpreted as angles in radians instead of degrees.

SplitChannel and SplitChanneld moved from transforms.post to transforms.utility

Removed

Support of PyTorch 1.4

Fixed

Enhanced the Dice related loss functions for stability and flexibility

Sliding window inference memory and device issues

Revised transforms:

Normalize intensity datatype and normalizer types

Padding modes for zoom

Crop returns coordinates

Select items transform

Weighted patch sampling

Option to keep aspect ratio for zoom

Various CI/CD issues

Source code(tar.gz)
Source code(zip)
0.3.0(Oct 5, 2020)
Added

Overview document for feature highlights in v0.3.0

Automatic mixed precision support

Multi-node, multi-GPU data parallel model training support

3 new evaluation metric functions

11 new network layers and blocks

6 new network architectures

14 new transforms, including an I/O adaptor

Cross validation module for DecathlonDataset

Smart Cache module in dataset

monai.optimizers module

monai.csrc module

Experimental feature of ImageReader using ITK, Nibabel, Numpy, Pillow (PIL Fork)

Experimental feature of differentiable image resampling in C++/CUDA

Ensemble evaluator module

GAN trainer module

Initial cross-platform CI environment for C++/CUDA code

Code style enforcement now includes isort and clang-format

Progress bar with tqdm

Changed

Now fully compatible with PyTorch 1.6

Base Docker image upgraded to nvcr.io/nvidia/pytorch:20.08-py3 from nvcr.io/nvidia/pytorch:20.03-py3

Code contributions now require signing off on the Developer Certificate of Origin (DCO)

Major work in type hinting finished

Remote datasets migrated to Open Data on AWS

Optionally depend on PyTorch-Ignite v0.4.2 instead of v0.3.0

Optionally depend on torchvision, ITK

Enhanced CI tests with 8 new testing environments

Removed

MONAI/examples folder (relocated into Project-MONAI/tutorials)

MONAI/research folder (relocated to Project-MONAI/research-contributions)

Fixed

dense_patch_slices incorrect indexing

Data type issue in GeneralizedWassersteinDiceLoss

ZipDataset return value inconsistencies

sliding_window_inference indexing and device issues

importing monai modules may cause namespace pollution

Random data splits issue in DecathlonDataset

Issue of randomising a Compose transform

Various issues in function type hints

Typos in docstring and documentation

PersistentDataset issue with existing file folder

Filename issue in the output writers

Source code(tar.gz)
Source code(zip)
0.2.0(Jul 7, 2020)
Added

Overview document for feature highlights in v0.2.0

Type hints and static type analysis support

MONAI/research folder

monai.engine.workflow APIs for supervised training

monai.inferers APIs for validation and inference

7 new tutorials and examples

3 new loss functions

4 new event handlers

8 new layers, blocks, and networks

12 new transforms, including post-processing transforms

monai.apps.datasets APIs, including MedNISTDataset and DecathlonDataset

Persistent caching, ZipDataset, and ArrayDataset in monai.data

Cross-platform CI tests supporting multiple Python versions

Optional import mechanism

Experimental features for third-party transforms integration

Changed

For more details please visit the project wiki

Core modules now require numpy >= 1.17

Categorized monai.transforms modules into crop and pad, intensity, IO, post-processing, spatial, and utility

Most transforms are now implemented with PyTorch native APIs

Code style enforcement and automated formatting workflows now use autopep8 and black

Base Docker image upgraded to nvcr.io/nvidia/pytorch:20.03-py3 from nvcr.io/nvidia/pytorch:19.10-py3

Enhanced local testing tools

Documentation website domain changed to https://docs.monai.io

Removed

Support of Python < 3.6

Automatic installation of optional dependencies including pytorch-ignite, nibabel, tensorboard, pillow, scipy, scikit-image

Fixed

Various issues in type and argument names consistency

Various issues in docstring and documentation site

Various issues in unit and integration tests

Various issues in examples and notebooks

Source code(tar.gz)
Source code(zip)
0.1.0(Apr 21, 2020)
Added

Public alpha source code under the Apache 2.0 license (highlights)

Various tutorials and examples

Medical image classification and segmentation workflows

Spacing/orientation-aware preprocessing with CPU/GPU and caching

Flexible workflows with external engines including PyTorch Ignite and Lightning

Various GitHub Actions

CI/CD pipelines via self-hosted runners

Documentation publishing via readthedocs.org

PyPI package publishing

Contributing guidelines

A project logo and badges

Source code(tar.gz)
Source code(zip)