File size: 12,143 Bytes
7ca1baf 9ed253c 09ebdbd 7ca1baf 3d17d70 28ec6a4 caa599e 3757b70 22317c0 caa599e f3fc250 16bb7f0 f3fc250 d6e2545 ec2329f 355b4c2 4f45167 692e8fc 4f45167 355b4c2 16bb7f0 99b0dd6 355b4c2 4f45167 e87edd2 4f45167 355b4c2 7d11710 7bd31a0 99b0dd6 7d11710 ec2329f 16bb7f0 355b4c2 16bb7f0 ec2329f 72ee961 ec2329f e3ac59a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 |
---
language:
- en
pipeline_tag: image-segmentation
tags:
- medical
library_name: fastMONAI
---
# Endometrial cancer segmentation
This repository contains weights and exported learner (encapsulates both the model architecture and its trained parameters) for a deep learning model designed to automate the segmentation of endometrial cancer on MR images.
Our VIBE model utilizes a Residual U-Net architecture, trained on data derived from the study [Automated segmentation of endometrial cancer on MR images using deep learning](https://link.springer.com/content/pdf/10.1038/s41598-020-80068-9.pdf).
The primary objective of this repository is to reproduce the results reported in the study and to integrate this model into research PACS (see [Results for VIBE](#results-for-vibe) section).
In addition, we have looked at improving the segmentation performance using multi-sequence MR images (T2w, VIBE, and ADC) (see [Results for multi-sequence (T2, VIBE, and ADC)](#results-for-multi-sequence-t2-vibe-and-adc) section).
## Requirements
Last checked and validated with fastMONAI version 0.3.9. Please ensure that you have the correct version of fastMONAI installed to guarantee the correct operation of the model.
## Usage
The source code for training the model and running inference on your own data is available at: https://github.com/MMIV-ML/fastMONAI/tree/master/research/endometrial_cancer.
Test our model live with the Gradio app for VIBE on [Hugging Face Spaces](https://skaliy-endometrial-cancer-segmentation-app.hf.space).
## Results for VIBE
Note that our results are not directly comparable with the results reported in [study](https://link.springer.com/content/pdf/10.1038/s41598-020-80068-9.pdf), as we opted to use the test set for validation to allocate more data to training. Unlike the approach detailed in the study, we refrained from post-processing steps, such as retaining only the largest object.
Predictions from new test cases indicate that this method could occasionally eliminate the tumor.
Below is the box plot showcasing predictions on the validation set:
The results from the validation set are also presented in the table below:
| | subject_id | tumor_vol | inter_rater | r1_ml | r2_ml | n_components |
|---:|-------------:|------------:|--------------:|---------:|-----------:|---------------:|
| 0 | 29 | 4.16 | 0.201835 | 0.806382 | 0.00623053 | 3 |
| 1 | 32 | 8 | 0.684142 | 0.293306 | 0.209449 | 4 |
| 2 | 36 | 19.06 | 0.92875 | 0.793055 | 0.784799 | 2 |
| 3 | 47 | 11.01 | 0.944209 | 0.900945 | 0.898409 | 2 |
| 4 | 50 | 6.26 | 0.722867 | 0.614357 | 0.624832 | 1 |
| 5 | 65 | 13.09 | 0.930613 | 0.879279 | 0.850546 | 2 |
| 6 | 67 | 3.71 | 0.943498 | 0.887189 | 0.878163 | 2 |
| 7 | 75 | 7.16 | 0.263539 | 0.774237 | 0.266619 | 2 |
| 8 | 86 | 7.04 | 0.842577 | 0.821208 | 0.798148 | 1 |
| 9 | 135 | 8.1 | 0.839964 | 0.758176 | 0.680348 | 2 |
| 10 | 140 | 19.78 | 0.895506 | 0.936177 | 0.874019 | 4 |
| 11 | 164 | 16.98 | 0.905008 | 0.923559 | 0.887268 | 1 |
| 12 | 246 | 6.59 | 0.899448 | 0.895311 | 0.860322 | 3 |
| 13 | 255 | 36.22 | 0.955784 | 0.927517 | 0.921816 | 6 |
| 14 | 343 | 0.69 | 0.528261 | 0.840237 | 0.600751 | 4 |
| 15 | 349 | 2.96 | 0.912664 | 0.828181 | 0.778983 | 1 |
| 16 | 367 | 1.02 | 0.0734848 | 0.391737 | 0.118035 | 1 |
| 17 | 370 | 10.82 | 0.953443 | 0.917094 | 0.908893 | 1 |
| 18 | 371 | 3.83 | 0.859781 | 0.684751 | 0.618114 | 1 |
| 19 | 375 | 11.67 | 0.911141 | 0.921079 | 0.91056 | 4 |
| 20 | 377 | 4.37 | 0.782994 | 0.712791 | 0.680165 | 1 |
| 21 | 381 | 7.63 | 0.89199 | 0.246428 | 0.238641 | 1 |
| 22 | 385 | 2.67 | 0.803215 | 0.641916 | 0.60169 | 1 |
| 23 | 395 | 0.68 | 0.770738 | 0.198273 | 0.236343 | 5 |
| 24 | 397 | 5.94 | 0.904544 | 0.882265 | 0.874036 | 3 |
| 25 | 409 | 11.86 | 0.944934 | 0.900727 | 0.900767 | 1 |
| 26 | 411 | 5.98 | 0.949977 | 0.933271 | 0.929499 | 1 |
| 27 | 425 | 0.91 | 0.802867 | 0.589069 | 0.545761 | 1 |
| 28 | 434 | 94.42 | 0.894601 | 0.590408 | 0.580585 | 1 |
| 29 | 531 | 22.08 | 0.89225 | 0.555066 | 0.505109 | 1 |
| 30 | 540 | 8.35 | 0.923702 | 0.855009 | 0.840958 | 1 |
<b>Median DSC</b>: 0.8946, 0.8212, 0.779
Prediction on a new subject in the research PACS:
## Results for multi-sequence (T2, VIBE, and ADC)
The box plot of the predictions on the validation set:
The results from the validation set are also presented in the table below:
| | subject_id | tumor_vol | inter_rater | r1_ml | r2_ml | n_components |
|---:|-------------:|------------:|--------------:|---------:|----------:|---------------:|
| 0 | 29 | 4.16 | 0.201835 | 0.859937 | 0.148586 | 4 |
| 1 | 32 | 8 | 0.684142 | 0.662779 | 0.515479 | 10 |
| 2 | 36 | 19.06 | 0.92875 | 0.902343 | 0.888306 | 1 |
| 3 | 47 | 11.01 | 0.944209 | 0.907344 | 0.907 | 3 |
| 4 | 50 | 6.26 | 0.722867 | 0.581594 | 0.540991 | 5 |
| 5 | 65 | 13.09 | 0.930613 | 0.889782 | 0.862255 | 4 |
| 6 | 67 | 3.71 | 0.943498 | 0.851658 | 0.842331 | 2 |
| 7 | 75 | 7.16 | 0.263539 | 0.750551 | 0.205457 | 2 |
| 8 | 86 | 7.04 | 0.842577 | 0.87216 | 0.81374 | 1 |
| 9 | 135 | 8.1 | 0.839964 | 0.80436 | 0.747164 | 1 |
| 10 | 140 | 19.78 | 0.895506 | 0.907457 | 0.852548 | 1 |
| 11 | 164 | 16.98 | 0.905008 | 0.92533 | 0.893135 | 2 |
| 12 | 246 | 6.59 | 0.899448 | 0.906569 | 0.852195 | 5 |
| 13 | 255 | 36.22 | 0.955784 | 0.924517 | 0.927624 | 2 |
| 14 | 343 | 0.69 | 0.528261 | 0.868251 | 0.457711 | 3 |
| 15 | 349 | 2.96 | 0.912664 | 0.85214 | 0.819898 | 1 |
| 16 | 367 | 1.02 | 0.0734848 | 0.383455 | 0.0891463 | 3 |
| 17 | 370 | 10.82 | 0.953443 | 0.916154 | 0.911768 | 2 |
| 18 | 371 | 3.83 | 0.859781 | 0.593136 | 0.565848 | 8 |
| 19 | 375 | 11.67 | 0.911141 | 0.898501 | 0.910147 | 3 |
| 20 | 377 | 4.37 | 0.782994 | 0.713798 | 0.646684 | 3 |
| 21 | 381 | 7.63 | 0.89199 | 0.4375 | 0.430847 | 1 |
| 22 | 385 | 2.67 | 0.803215 | 0.688608 | 0.624595 | 1 |
| 23 | 395 | 0.68 | 0.770738 | 0.385992 | 0.43154 | 2 |
| 24 | 397 | 5.94 | 0.904544 | 0.868022 | 0.850653 | 6 |
| 25 | 409 | 11.86 | 0.944934 | 0.83407 | 0.833206 | 5 |
| 26 | 411 | 5.98 | 0.949977 | 0.867137 | 0.866112 | 1 |
| 27 | 425 | 0.91 | 0.802867 | 0.557732 | 0.475499 | 3 |
| 28 | 434 | 94.42 | 0.894601 | 0.618916 | 0.605596 | 6 |
| 29 | 531 | 22.08 | 0.89225 | 0.349648 | 0.319533 | 1 |
| 30 | 540 | 8.35 | 0.923702 | 0.890343 | 0.88052 | 1 |
<b>Median DSC</b>: 0.8946, 0.8521, 0.8137
## Results for multi-sequence (T2, VIBE, and ADC) with extra training data (n=54)
Need to run cross-validation to make a better comparison.
| | subject_id | tumor_vol | inter_rater | r1_ml | r2_ml | n_components |
|---:|-------------:|------------:|--------------:|---------:|----------:|---------------:|
| 0 | 29 | 4.16 | 0.201835 | 0.836437 | 0.0599303 | 2 |
| 1 | 32 | 8 | 0.684142 | 0.65186 | 0.503093 | 11 |
| 2 | 36 | 19.06 | 0.92875 | 0.876779 | 0.862773 | 3 |
| 3 | 47 | 11.01 | 0.944209 | 0.914218 | 0.911429 | 4 |
| 4 | 50 | 6.26 | 0.722867 | 0.667869 | 0.60398 | 1 |
| 5 | 65 | 13.09 | 0.930613 | 0.88374 | 0.859066 | 1 |
| 6 | 67 | 3.71 | 0.943498 | 0.861391 | 0.851904 | 1 |
| 7 | 75 | 7.16 | 0.263539 | 0.769195 | 0.236445 | 4 |
| 8 | 86 | 7.04 | 0.842577 | 0.848937 | 0.80314 | 3 |
| 9 | 135 | 8.1 | 0.839964 | 0.810392 | 0.732383 | 1 |
| 10 | 140 | 19.78 | 0.895506 | 0.92261 | 0.865316 | 1 |
| 11 | 164 | 16.98 | 0.905008 | 0.923593 | 0.879799 | 5 |
| 12 | 246 | 6.59 | 0.899448 | 0.919342 | 0.864234 | 1 |
| 13 | 255 | 36.22 | 0.955784 | 0.939234 | 0.938806 | 2 |
| 14 | 343 | 0.69 | 0.528261 | 0.839357 | 0.448649 | 5 |
| 15 | 349 | 2.96 | 0.912664 | 0.877018 | 0.839009 | 1 |
| 16 | 367 | 1.02 | 0.0734848 | 0.255149 | 0.0615073 | 1 |
| 17 | 370 | 10.82 | 0.953443 | 0.916431 | 0.907043 | 18 |
| 18 | 371 | 3.83 | 0.859781 | 0.508698 | 0.475138 | 1 |
| 19 | 375 | 11.67 | 0.911141 | 0.90593 | 0.910805 | 1 |
| 20 | 377 | 4.37 | 0.782994 | 0.622583 | 0.598939 | 4 |
| 21 | 381 | 7.63 | 0.89199 | 0.392978 | 0.381061 | 1 |
| 22 | 385 | 2.67 | 0.803215 | 0.666327 | 0.583576 | 2 |
| 23 | 395 | 0.68 | 0.770738 | 0.53442 | 0.54433 | 4 |
| 24 | 397 | 5.94 | 0.904544 | 0.867964 | 0.868074 | 5 |
| 25 | 409 | 11.86 | 0.944934 | 0.826939 | 0.827658 | 5 |
| 26 | 411 | 5.98 | 0.949977 | 0.786394 | 0.796158 | 1 |
| 27 | 425 | 0.91 | 0.802867 | 0.508261 | 0.43545 | 1 |
| 28 | 434 | 94.42 | 0.894601 | 0.77102 | 0.758085 | 2 |
| 29 | 531 | 22.08 | 0.89225 | 0.271076 | 0.253303 | 1 |
| 30 | 540 | 8.35 | 0.923702 | 0.898613 | 0.890637 | 1 |
<b>Median DSC</b>: 0.8946, 0.8364, 0.7962
## Support and Contribution
For any issues related to the model or the source code, please open an issue in the corresponding GitHub repository. Contributions to the code or the model are welcome and should be proposed through a pull request. |