Relevance maps: A weakly supervised segmentation method for 3D brain tumours in MRIs
With the increased reliance on medical imaging, Deep convolutional neural networks (CNNs) have become an essential tool in the medical imaging-based computer-aided diagnostic pipelines. However, training accurate and reliable classification models often require large fine-grained annotated datasets....
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2022-12-01
|
| Series: | Frontiers in Radiology |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fradi.2022.1061402/full |
| _version_ | 1797979752871493632 |
|---|---|
| author | Sajith Rajapaksa Sajith Rajapaksa Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati |
| author_facet | Sajith Rajapaksa Sajith Rajapaksa Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati |
| author_sort | Sajith Rajapaksa |
| collection | DOAJ |
| description | With the increased reliance on medical imaging, Deep convolutional neural networks (CNNs) have become an essential tool in the medical imaging-based computer-aided diagnostic pipelines. However, training accurate and reliable classification models often require large fine-grained annotated datasets. To alleviate this, weakly-supervised methods can be used to obtain local information such as region of interest from global labels. This work proposes a weakly-supervised pipeline to extract Relevance Maps of medical images from pre-trained 3D classification models using localized perturbations. The extracted Relevance Map describes a given region’s importance to the classification model and produces the segmentation for the region. Furthermore, we propose a novel optimal perturbation generation method that exploits 3D superpixels to find the most relevant area for a given classification using U-net architecture. This model is trained with perturbation loss, which maximizes the difference between unperturbed and perturbed predictions. We validated the effectiveness of our methodology by applying it to the segmentation of Glioma brain tumours in MRI scans using only classification labels for glioma type. The proposed method outperforms existing methods in both Dice Similarity Coefficient for segmentation and resolution for visualizations. |
| first_indexed | 2024-04-11T05:43:40Z |
| format | Article |
| id | doaj.art-9c0b149e2078499f95a75b75d9b867cf |
| institution | Directory Open Access Journal |
| issn | 2673-8740 |
| language | English |
| last_indexed | 2024-04-11T05:43:40Z |
| publishDate | 2022-12-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Radiology |
| spelling | doaj.art-9c0b149e2078499f95a75b75d9b867cf2022-12-22T04:42:20ZengFrontiers Media S.A.Frontiers in Radiology2673-87402022-12-01210.3389/fradi.2022.10614021061402Relevance maps: A weakly supervised segmentation method for 3D brain tumours in MRIsSajith Rajapaksa0Sajith Rajapaksa1Farzad Khalvati2Farzad Khalvati3Farzad Khalvati4Farzad Khalvati5Farzad Khalvati6Farzad Khalvati7Farzad Khalvati8Neurosciences and Mental Health, The Hospital for Sick Children, Toronto, ON, CanadaInstitute of Medical Science, University of Toronto, Toronto, ON, CanadaNeurosciences and Mental Health, The Hospital for Sick Children, Toronto, ON, CanadaDepartment of Diagnostic Imaging, The Hospital for Sick Children, Toronto, ON, CanadaInstitute of Medical Science, University of Toronto, Toronto, ON, CanadaDepartment of Medical Imaging, University of Toronto, Toronto, ON, CanadaDepartment of Mechanical and Industrial Engineering, University of Toronto, Toronto, ON, CanadaDepartment of Computer Science, University of Toronto, Toronto, ON, CanadaVector Institute, Toronto, ON, CanadaWith the increased reliance on medical imaging, Deep convolutional neural networks (CNNs) have become an essential tool in the medical imaging-based computer-aided diagnostic pipelines. However, training accurate and reliable classification models often require large fine-grained annotated datasets. To alleviate this, weakly-supervised methods can be used to obtain local information such as region of interest from global labels. This work proposes a weakly-supervised pipeline to extract Relevance Maps of medical images from pre-trained 3D classification models using localized perturbations. The extracted Relevance Map describes a given region’s importance to the classification model and produces the segmentation for the region. Furthermore, we propose a novel optimal perturbation generation method that exploits 3D superpixels to find the most relevant area for a given classification using U-net architecture. This model is trained with perturbation loss, which maximizes the difference between unperturbed and perturbed predictions. We validated the effectiveness of our methodology by applying it to the segmentation of Glioma brain tumours in MRI scans using only classification labels for glioma type. The proposed method outperforms existing methods in both Dice Similarity Coefficient for segmentation and resolution for visualizations.https://www.frontiersin.org/articles/10.3389/fradi.2022.1061402/fullweakly supervisedsuperpixelsMRIbrain tumourssegmentationexplainability |
| spellingShingle | Sajith Rajapaksa Sajith Rajapaksa Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati Farzad Khalvati Relevance maps: A weakly supervised segmentation method for 3D brain tumours in MRIs Frontiers in Radiology weakly supervised superpixels MRI brain tumours segmentation explainability |
| title | Relevance maps: A weakly supervised segmentation method for 3D brain tumours in MRIs |
| title_full | Relevance maps: A weakly supervised segmentation method for 3D brain tumours in MRIs |
| title_fullStr | Relevance maps: A weakly supervised segmentation method for 3D brain tumours in MRIs |
| title_full_unstemmed | Relevance maps: A weakly supervised segmentation method for 3D brain tumours in MRIs |
| title_short | Relevance maps: A weakly supervised segmentation method for 3D brain tumours in MRIs |
| title_sort | relevance maps a weakly supervised segmentation method for 3d brain tumours in mris |
| topic | weakly supervised superpixels MRI brain tumours segmentation explainability |
| url | https://www.frontiersin.org/articles/10.3389/fradi.2022.1061402/full |
| work_keys_str_mv | AT sajithrajapaksa relevancemapsaweaklysupervisedsegmentationmethodfor3dbraintumoursinmris AT sajithrajapaksa relevancemapsaweaklysupervisedsegmentationmethodfor3dbraintumoursinmris AT farzadkhalvati relevancemapsaweaklysupervisedsegmentationmethodfor3dbraintumoursinmris AT farzadkhalvati relevancemapsaweaklysupervisedsegmentationmethodfor3dbraintumoursinmris AT farzadkhalvati relevancemapsaweaklysupervisedsegmentationmethodfor3dbraintumoursinmris AT farzadkhalvati relevancemapsaweaklysupervisedsegmentationmethodfor3dbraintumoursinmris AT farzadkhalvati relevancemapsaweaklysupervisedsegmentationmethodfor3dbraintumoursinmris AT farzadkhalvati relevancemapsaweaklysupervisedsegmentationmethodfor3dbraintumoursinmris AT farzadkhalvati relevancemapsaweaklysupervisedsegmentationmethodfor3dbraintumoursinmris |