Join Our Discord (630+ Members)

Prostate MRI Zonal Segmentation

A pre-trained model for volumetric (3D) segmentation of the prostate from MRI images

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

Subscribe or Contribute

A pre-trained model for volumetric (3D) segmentation of the prostate from MRI images

This model was trained with the UNet architecture [1] and is used for 3D volumetric segmentation of the anatomical prostate zones on T2w MRI images. The segmentation of the anatomical regions is formulated as a voxel-wise classification. Each voxel is classified as either central gland (1), peripheral zone (2), or background (0). The model is optimized using a gradient descent method that minimizes the focal soft-dice loss between the predicted mask and the actual segmentation.

Data

The model was trained in the prostate158 training data, which is available at https://doi.org/10.5281/zenodo.6481141 . Only T2w images were used for this task.

Preprocessing

MRI images in the prostate158 dataset were preprocessed, including center cropping and resampling. When applying the model to new data, this preprocessing should be repeated.

Center cropping

T2w images were acquired with a voxel spacing of 0.47 x 0.47 x 3 mm and an axial FOV size of 180 x 180 mm. However, the prostate rarely exceeds an axial diameter of 100 mm, and for zonal segmentation, the tissue surrounding the prostate is not of interest and only increases the image size and thus the computational cost. Center-cropping can reduce the image size without sacrificing information.

The script center_crop.py allows to reproduce center-cropping as performed in the prostate158 paper.

python scripts/center_crop.py --file_name path/to/t2_image --out_name cropped_t2

Resampling

DWI and ADC sequences in prostate158 were resampled to the orientation and voxel spacing of the T2w sequence. As the zonal segmentation uses T2w images, no additional resampling is nessecary. However, the training script will perform additonal resampling automatically.

Performance

The model achives the following performance on the prostate158 test dataset:

  • Rater 1
  • Rater 2
  • Metric
  • Transitional Zone
  • Peripheral Zone
  • Transitional Zone
  • Peripheral Zone

Dice Coefficient

0.877

0.754

0.875

0.730

Hausdorff Distance

18.3

22.8

17.5

33.2

Surface Distance

2.19

1.95

2.59

1.88

For more details, please see the original publication or official GitHub repository

System Configuration

The model was trained for 100 epochs on a workstaion with a single Nvidia RTX 3080 GPU. This takes approximatly 8 hours.

Limitations (Optional)

This training and inference pipeline was developed for research purposes only. This research use only software that has not been cleared or approved by FDA or any regulatory agency. The model is for research/developmental purposes only and cannot be used directly for clinical procedures.

Citation Info (Optional)

@article{ADAMS2022105817,
title = {Prostate158 - An expert-annotated 3T MRI dataset and algorithm for prostate cancer detection},
journal = {Computers in Biology and Medicine},
volume = {148},
pages = {105817},
year = {2022},
issn = {0010-4825},
doi = {https://doi.org/10.1016/j.compbiomed.2022.105817},
url = {https://www.sciencedirect.com/science/article/pii/S0010482522005789},
author = {Lisa C. Adams and Marcus R. Makowski and Günther Engel and Maximilian Rattunde and Felix Busch and Patrick Asbach and Stefan M. Niehues and Shankeeth Vinayahalingam and Bram {van Ginneken} and Geert Litjens and Keno K. Bressem},
keywords = {Prostate cancer, Deep learning, Machine learning, Artificial intelligence, Magnetic resonance imaging, Biparametric prostate MRI}
}

References

[1] Sakinis, Tomas, et al. “Interactive segmentation of medical images through fully convolutional neural networks.” arXiv preprint arXiv:1903.08205 (2019).

License

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

http://www.apache.org/licenses/LICENSE-2.0

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.

Related Posts

6lack AI Voice

6lack AI Voice

Experience the groundbreaking sound of AI-based music with our collection of songs by 6lack.

Endoscopic Inbody Classification

Endoscopic Inbody Classification

A pre-trained binary classification model for endoscopic inbody classification task A pre-trained model for the endoscopic inbody classification task and trained using the SEResNet50 structure, whose details can be found in [1].

Tate Mcrae AI Voice

Tate Mcrae AI Voice

Discover a unique collection of songs by AI-Tate Mcrae, featuring vocals generated by a community of AI enthusiasts.