Fidelity imposed network edit (FINE) for solving ill-posed image reconstruction
Deep learning (DL) is increasingly used to solve ill-posed inverse problems in medical imaging, such as reconstruction from noisy and/or incomplete data, as DL offers advantages over conventional methods that rely on explicit image features and hand engineered priors. However, supervised DL-based me...
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2020-05-01
|
| Series: | NeuroImage |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1053811920300665 |
| _version_ | 1819075420058812416 |
|---|---|
| author | Jinwei Zhang Zhe Liu Shun Zhang Hang Zhang Pascal Spincemaille Thanh D. Nguyen Mert R. Sabuncu Yi Wang |
| author_facet | Jinwei Zhang Zhe Liu Shun Zhang Hang Zhang Pascal Spincemaille Thanh D. Nguyen Mert R. Sabuncu Yi Wang |
| author_sort | Jinwei Zhang |
| collection | DOAJ |
| description | Deep learning (DL) is increasingly used to solve ill-posed inverse problems in medical imaging, such as reconstruction from noisy and/or incomplete data, as DL offers advantages over conventional methods that rely on explicit image features and hand engineered priors. However, supervised DL-based methods may achieve poor performance when the test data deviates from the training data, for example, when it has pathologies not encountered in the training data. Furthermore, DL-based image reconstructions do not always incorporate the underlying forward physical model, which may improve performance. Therefore, in this work we introduce a novel approach, called fidelity imposed network edit (FINE), which modifies the weights of a pre-trained reconstruction network for each case in the testing dataset. This is achieved by minimizing an unsupervised fidelity loss function that is based on the forward physical model. FINE is applied to two important inverse problems in neuroimaging: quantitative susceptibility mapping (QSM) and under-sampled image reconstruction in MRI. Our experiments demonstrate that FINE can improve reconstruction accuracy. |
| first_indexed | 2024-12-21T18:25:07Z |
| format | Article |
| id | doaj.art-4a304430a53a4259bde40c97529d9fbc |
| institution | Directory Open Access Journal |
| issn | 1095-9572 |
| language | English |
| last_indexed | 2024-12-21T18:25:07Z |
| publishDate | 2020-05-01 |
| publisher | Elsevier |
| record_format | Article |
| series | NeuroImage |
| spelling | doaj.art-4a304430a53a4259bde40c97529d9fbc2022-12-21T18:54:27ZengElsevierNeuroImage1095-95722020-05-01211116579Fidelity imposed network edit (FINE) for solving ill-posed image reconstructionJinwei Zhang0Zhe Liu1Shun Zhang2Hang Zhang3Pascal Spincemaille4Thanh D. Nguyen5Mert R. Sabuncu6Yi Wang7Department of Biomedical Engineering, Cornell University, Ithaca, NY, USA; Department of Radiology, Weill Medical College of Cornell University, New York, NY, USADepartment of Biomedical Engineering, Cornell University, Ithaca, NY, USA; Department of Radiology, Weill Medical College of Cornell University, New York, NY, USADepartment of Radiology, Weill Medical College of Cornell University, New York, NY, USADepartment of Radiology, Weill Medical College of Cornell University, New York, NY, USA; Department of Electrical and Computer Engineering, Cornell University, Ithaca, NY, USADepartment of Radiology, Weill Medical College of Cornell University, New York, NY, USADepartment of Radiology, Weill Medical College of Cornell University, New York, NY, USADepartment of Biomedical Engineering, Cornell University, Ithaca, NY, USA; Department of Radiology, Weill Medical College of Cornell University, New York, NY, USA; Department of Electrical and Computer Engineering, Cornell University, Ithaca, NY, USADepartment of Biomedical Engineering, Cornell University, Ithaca, NY, USA; Department of Radiology, Weill Medical College of Cornell University, New York, NY, USA; Corresponding author. Department of Radiology, Weill Cornell Medicine, 515 E 71th Street, New York, NY, 10021, USA.Deep learning (DL) is increasingly used to solve ill-posed inverse problems in medical imaging, such as reconstruction from noisy and/or incomplete data, as DL offers advantages over conventional methods that rely on explicit image features and hand engineered priors. However, supervised DL-based methods may achieve poor performance when the test data deviates from the training data, for example, when it has pathologies not encountered in the training data. Furthermore, DL-based image reconstructions do not always incorporate the underlying forward physical model, which may improve performance. Therefore, in this work we introduce a novel approach, called fidelity imposed network edit (FINE), which modifies the weights of a pre-trained reconstruction network for each case in the testing dataset. This is achieved by minimizing an unsupervised fidelity loss function that is based on the forward physical model. FINE is applied to two important inverse problems in neuroimaging: quantitative susceptibility mapping (QSM) and under-sampled image reconstruction in MRI. Our experiments demonstrate that FINE can improve reconstruction accuracy.http://www.sciencedirect.com/science/article/pii/S1053811920300665Data fidelityDeep learningInverse problemUnder-sampled image reconstructionQuantitative susceptibility mapping |
| spellingShingle | Jinwei Zhang Zhe Liu Shun Zhang Hang Zhang Pascal Spincemaille Thanh D. Nguyen Mert R. Sabuncu Yi Wang Fidelity imposed network edit (FINE) for solving ill-posed image reconstruction NeuroImage Data fidelity Deep learning Inverse problem Under-sampled image reconstruction Quantitative susceptibility mapping |
| title | Fidelity imposed network edit (FINE) for solving ill-posed image reconstruction |
| title_full | Fidelity imposed network edit (FINE) for solving ill-posed image reconstruction |
| title_fullStr | Fidelity imposed network edit (FINE) for solving ill-posed image reconstruction |
| title_full_unstemmed | Fidelity imposed network edit (FINE) for solving ill-posed image reconstruction |
| title_short | Fidelity imposed network edit (FINE) for solving ill-posed image reconstruction |
| title_sort | fidelity imposed network edit fine for solving ill posed image reconstruction |
| topic | Data fidelity Deep learning Inverse problem Under-sampled image reconstruction Quantitative susceptibility mapping |
| url | http://www.sciencedirect.com/science/article/pii/S1053811920300665 |
| work_keys_str_mv | AT jinweizhang fidelityimposednetworkeditfineforsolvingillposedimagereconstruction AT zheliu fidelityimposednetworkeditfineforsolvingillposedimagereconstruction AT shunzhang fidelityimposednetworkeditfineforsolvingillposedimagereconstruction AT hangzhang fidelityimposednetworkeditfineforsolvingillposedimagereconstruction AT pascalspincemaille fidelityimposednetworkeditfineforsolvingillposedimagereconstruction AT thanhdnguyen fidelityimposednetworkeditfineforsolvingillposedimagereconstruction AT mertrsabuncu fidelityimposednetworkeditfineforsolvingillposedimagereconstruction AT yiwang fidelityimposednetworkeditfineforsolvingillposedimagereconstruction |