Join Our Discord (750+ Members)

Nuclei Cell Segmentation With User Clicks Annotation

A pre-trained model for segmenting nuclei cells with user clicks/interactions

A pre-trained model for segmenting nuclei cells with user clicks/interactions

nuclick nuclick nuclick

This model is trained using BasicUNet over ConSeP dataset.


The training dataset is from _fac/tia/data/hovernet

unzip -q


After downloading this dataset , python script from scripts folder can be used to preprocess and generate the final dataset for training.

python scripts/ --input /path/to/data/CoNSeP --output /path/to/data/CoNSePNuclei

After generating the output files, please modify the dataset_dir parameter specified in configs/train.json and configs/inference.json to reflect the output folder which contains new dataset.json.

Class values in dataset are

  • 1 = other
  • 2 = inflammatory
  • 3 = healthy epithelial
  • 4 = dysplastic/malignant epithelial
  • 5 = fibroblast
  • 6 = muscle
  • 7 = endothelial

As part of pre-processing, the following steps are executed.

  • Crop and Extract each nuclei Image + Label (128x128) based on the centroid given in the dataset.
  • Combine classes 3 & 4 into the epithelial class and 5,6 & 7 into the spindle-shaped class.
  • Update the label index for the target nuclei based on the class value
  • Other cells which are part of the patch are modified to have label idx = 255

Example dataset.json

  "training": [
      "image": "/workspace/data/CoNSePNuclei/Train/Images/train_1_3_0001.png",
      "label": "/workspace/data/CoNSePNuclei/Train/Labels/train_1_3_0001.png",
      "nuclei_id": 1,
      "mask_value": 3,
      "centroid": [
  "validation": [
      "image": "/workspace/data/CoNSePNuclei/Test/Images/test_1_3_0001.png",
      "label": "/workspace/data/CoNSePNuclei/Test/Labels/test_1_3_0001.png",
      "nuclei_id": 1,
      "mask_value": 3,
      "centroid": [

Training Configuration

The training was performed with the following:

  • GPU: at least 12GB of GPU memory
  • Actual Model Input: 5 x 128 x 128
  • AMP: True
  • Optimizer: Adam
  • Learning Rate: 1e-4
  • Loss: DiceLoss

Memory Consumption

  • Dataset Manager: CacheDataset
  • Data Size: 13,136 PNG images
  • Cache Rate: 1.0
  • Single GPU - System RAM Usage: 4.7G

Memory Consumption Warning

If you face memory issues with CacheDataset, you can either switch to a regular Dataset class or lower the caching rate cache_rate in the configurations within range [0, 1] to minimize the System RAM requirements.


5 channels - 3 RGB channels - +ve signal channel (this nuclei) - -ve signal channel (other nuclei)


2 channels - 0 = Background - 1 = Nuclei


This model achieves the following Dice score on the validation data provided as part of the dataset:

  • Train Dice score = 0.89
  • Validation Dice score = 0.85

Training Loss and Dice

A graph showing the training Loss and Dice over 50 epochs.

Validation Dice

A graph showing the validation mean Dice over 50 epochs.

MONAI Bundle Commands

In addition to the Pythonic APIs, a few command line interfaces (CLI) are provided to interact with the bundle. The CLI supports flexible use cases, such as overriding configs at runtime and predefining arguments in a file.

For more details usage instructions, visit the MONAI Bundle Configuration Page .

Execute training:

python -m monai.bundle run --config_file configs/train.json

Please note that if the default dataset path is not modified with the actual path in the bundle config files, you can also override it by using --dataset_dir :

python -m monai.bundle run --config_file configs/train.json --dataset_dir <actual dataset path>

Override the train config to execute multi-GPU training:

torchrun --standalone --nnodes=1 --nproc_per_node=2 -m monai.bundle run --config_file "['configs/train.json','configs/multi_gpu_train.json']"

Please note that the distributed training-related options depend on the actual running environment; thus, users may need to remove --standalone , modify --nnodes , or do some other necessary changes according to the machine used. For more details, please refer to pytorch’s official tutorial .

Override the train config to execute evaluation with the trained model:

python -m monai.bundle run --config_file "['configs/train.json','configs/evaluate.json']"

Override the train config and evaluate config to execute multi-GPU evaluation:

torchrun --standalone --nnodes=1 --nproc_per_node=2 -m monai.bundle run --config_file "['configs/train.json','configs/evaluate.json','configs/multi_gpu_evaluate.json']"

Execute inference:

python -m monai.bundle run --config_file configs/inference.json


[1] Koohbanani, Navid Alemi, et al. “NuClick: a deep learning framework for interactive segmentation of microscopic images.” Medical Image Analysis 65 (2020): 101771. .

[2] S. Graham, Q. D. Vu, S. E. A. Raza, A. Azam, Y-W. Tsang, J. T. Kwak and N. Rajpoot. “HoVer-Net: Simultaneous Segmentation and Classification of Nuclei in Multi-Tissue Histology Images.” Medical Image Analysis, Sept. 2019. [ doi ]

[3] NuClick PyTorch Implementation


Copyright (c) MONAI Consortium

Licensed under the Apache License, Version 2.0 (the “License”); you may not use this file except in compliance with the License. You may obtain a copy of the License at

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Download Close

Follow AI Models on Google News

An easy & free way to support AI Models is to follow our google news feed! More followers will help us reach a wider audience!

Google News: AI Models

Related Posts

Damon Albarn RVC Model AI Voice

Damon Albarn RVC Model AI Voice

Dive into the innovative world of AI-generated music with Damon Albarn’s latest collection.

Enrico Pucci (JoJo) RVC Model AI Voice

Enrico Pucci (JoJo) RVC Model AI Voice

Discover the unique sound of AI Enrico Pucci (JoJo) with our collection of songs created using VITS Retrieval based Voice Conversion tech.

Cohere - Aya

Cohere - Aya

C4AI’s research team introduces Aya, a state-of-the-art, open-source, multilingual AI model covering 101 languages.