Few-shot image segmentation for cross-institution male pelvic organs using registration-assisted prototypical learning
The ability to adapt medical image segmentation networks for a novel class such as an unseen anatomical or pathological structure, when only a few labelled examples of this class are available from local healthcare providers, is sought-after. This potentially addresses two widely recognised limitati...
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Internet publication |
| Language: | English |
| Published: |
arXiv
2022
|
| _version_ | 1797111905781809152 |
|---|---|
| author | Li, Y Fu, Y Yang, Q Min, Z Yan, W Huisman, H Barratt, D Prisacariu, V Hu, Y |
| author_facet | Li, Y Fu, Y Yang, Q Min, Z Yan, W Huisman, H Barratt, D Prisacariu, V Hu, Y |
| author_sort | Li, Y |
| collection | OXFORD |
| description | The ability to adapt medical image segmentation networks for a novel class such as an unseen anatomical or pathological structure, when only a few labelled examples of this class are available from local healthcare providers, is sought-after. This potentially addresses two widely recognised limitations in deploying modern deep learning models to clinical practice, expertise-and-labour-intensive labelling and cross-institution generalisation. This work presents the first 3D few-shot interclass segmentation network for medical images, using a labelled multi-institution dataset from prostate cancer patients with eight regions of interest. We propose an image alignment module registering the predicted segmentation of both query and support data, in a standard prototypical learning algorithm, to a reference atlas space. The built-in registration mechanism can effectively utilise the prior knowledge of consistent anatomy between subjects, regardless whether they are from the same institution or not. Experimental results demonstrated that the proposed registration-assisted prototypical learning significantly improved segmentation accuracy (p-values<0.01) on query data from a holdout institution, with varying availability of support data from multiple institutions. We also report the additional benefits of the proposed 3D networks with 75% fewer parameters and an arguably simpler implementation, compared with existing 2D few-shot approaches that segment 2D slices of volumetric medical images. |
| first_indexed | 2024-03-07T08:16:56Z |
| format | Internet publication |
| id | oxford-uuid:428c030e-5555-43d4-8dca-8c7116d2de70 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T08:16:56Z |
| publishDate | 2022 |
| publisher | arXiv |
| record_format | dspace |
| spelling | oxford-uuid:428c030e-5555-43d4-8dca-8c7116d2de702024-01-11T15:31:36ZFew-shot image segmentation for cross-institution male pelvic organs using registration-assisted prototypical learningInternet publicationhttp://purl.org/coar/resource_type/c_7ad9uuid:428c030e-5555-43d4-8dca-8c7116d2de70EnglishSymplectic ElementsarXiv2022Li, YFu, YYang, QMin, ZYan, WHuisman, HBarratt, DPrisacariu, VHu, YThe ability to adapt medical image segmentation networks for a novel class such as an unseen anatomical or pathological structure, when only a few labelled examples of this class are available from local healthcare providers, is sought-after. This potentially addresses two widely recognised limitations in deploying modern deep learning models to clinical practice, expertise-and-labour-intensive labelling and cross-institution generalisation. This work presents the first 3D few-shot interclass segmentation network for medical images, using a labelled multi-institution dataset from prostate cancer patients with eight regions of interest. We propose an image alignment module registering the predicted segmentation of both query and support data, in a standard prototypical learning algorithm, to a reference atlas space. The built-in registration mechanism can effectively utilise the prior knowledge of consistent anatomy between subjects, regardless whether they are from the same institution or not. Experimental results demonstrated that the proposed registration-assisted prototypical learning significantly improved segmentation accuracy (p-values<0.01) on query data from a holdout institution, with varying availability of support data from multiple institutions. We also report the additional benefits of the proposed 3D networks with 75% fewer parameters and an arguably simpler implementation, compared with existing 2D few-shot approaches that segment 2D slices of volumetric medical images. |
| spellingShingle | Li, Y Fu, Y Yang, Q Min, Z Yan, W Huisman, H Barratt, D Prisacariu, V Hu, Y Few-shot image segmentation for cross-institution male pelvic organs using registration-assisted prototypical learning |
| title | Few-shot image segmentation for cross-institution male pelvic organs using registration-assisted prototypical learning |
| title_full | Few-shot image segmentation for cross-institution male pelvic organs using registration-assisted prototypical learning |
| title_fullStr | Few-shot image segmentation for cross-institution male pelvic organs using registration-assisted prototypical learning |
| title_full_unstemmed | Few-shot image segmentation for cross-institution male pelvic organs using registration-assisted prototypical learning |
| title_short | Few-shot image segmentation for cross-institution male pelvic organs using registration-assisted prototypical learning |
| title_sort | few shot image segmentation for cross institution male pelvic organs using registration assisted prototypical learning |
| work_keys_str_mv | AT liy fewshotimagesegmentationforcrossinstitutionmalepelvicorgansusingregistrationassistedprototypicallearning AT fuy fewshotimagesegmentationforcrossinstitutionmalepelvicorgansusingregistrationassistedprototypicallearning AT yangq fewshotimagesegmentationforcrossinstitutionmalepelvicorgansusingregistrationassistedprototypicallearning AT minz fewshotimagesegmentationforcrossinstitutionmalepelvicorgansusingregistrationassistedprototypicallearning AT yanw fewshotimagesegmentationforcrossinstitutionmalepelvicorgansusingregistrationassistedprototypicallearning AT huismanh fewshotimagesegmentationforcrossinstitutionmalepelvicorgansusingregistrationassistedprototypicallearning AT barrattd fewshotimagesegmentationforcrossinstitutionmalepelvicorgansusingregistrationassistedprototypicallearning AT prisacariuv fewshotimagesegmentationforcrossinstitutionmalepelvicorgansusingregistrationassistedprototypicallearning AT huy fewshotimagesegmentationforcrossinstitutionmalepelvicorgansusingregistrationassistedprototypicallearning |