Octascope: A Lightweight Pre-Trained Model for Optical Coherence Tomography
Optical coherence tomography (OCT) imaging enables high resolution visualization of sub-surface tissue microstructures. However, OCT image analysis using deep learning is hampered by limited diverse training data to meet performance requirements and high inference latency for real-time applications....
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11113249/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849236307857899520 |
|---|---|
| author | Haoyang Cui Chen Wang Paul Calle Yunlong Liu Qinghao Zhang Sinaro Ly Justin Reynolds Feng Yan Ke Zhang Ronghao Liu Junyuan Liu Kar-Ming Fung Zhongxin Yu Ajay Jain Qinggong Tang Chongle Pan |
| author_facet | Haoyang Cui Chen Wang Paul Calle Yunlong Liu Qinghao Zhang Sinaro Ly Justin Reynolds Feng Yan Ke Zhang Ronghao Liu Junyuan Liu Kar-Ming Fung Zhongxin Yu Ajay Jain Qinggong Tang Chongle Pan |
| author_sort | Haoyang Cui |
| collection | DOAJ |
| description | Optical coherence tomography (OCT) imaging enables high resolution visualization of sub-surface tissue microstructures. However, OCT image analysis using deep learning is hampered by limited diverse training data to meet performance requirements and high inference latency for real-time applications. To address these challenges, we developed Octascope, a lightweight domain-specific convolutional neural network (CNN) - based model designed for OCT image analysis. Octascope was pre-trained using a curriculum learning approach, which involves sequential training, first on natural images (ImageNet), then on OCT images from retinal, abdominal, and renal tissues, to progressively acquire transferable knowledge. This multi-domain pre-training enables Octascope to generalize across varied tissue types. In two downstream tasks, Octascope demonstrated notable improvements in predictive accuracy compared to alternative approaches. In the epidural tissue detection task, our method surpassed single-task learning with fine-tuning by 9.13% and OCT-specific transfer learning by 5.95% in accuracy. Octascope outperformed VGG16 and ResNet50 by 5.36% and 6.66% in a retinal diagnosis task, respectively. In comparison to a Transformer-based OCT foundation model - RETFound, Octascope delivered 2 to 4.4 times faster inference speed with slightly better predictive accuracies in both downstream tasks. Octascope represented a significant advancement for OCT image analysis by providing an effective balance between computational efficiency and diagnostic accuracy for real-time clinical applications. |
| format | Article |
| id | doaj-art-21d4eec4f7cc4e239e8bf92d9acdfe25 |
| institution | Kabale University |
| issn | 2169-3536 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj-art-21d4eec4f7cc4e239e8bf92d9acdfe252025-08-20T04:02:18ZengIEEEIEEE Access2169-35362025-01-011313800513801910.1109/ACCESS.2025.359583811113249Octascope: A Lightweight Pre-Trained Model for Optical Coherence TomographyHaoyang Cui0https://orcid.org/0009-0009-8708-5556Chen Wang1https://orcid.org/0000-0003-4645-3227Paul Calle2https://orcid.org/0009-0000-1849-4481Yunlong Liu3Qinghao Zhang4Sinaro Ly5https://orcid.org/0009-0002-5269-9717Justin Reynolds6Feng Yan7https://orcid.org/0000-0001-9926-7554Ke Zhang8https://orcid.org/0000-0003-3194-2546Ronghao Liu9Junyuan Liu10Kar-Ming Fung11Zhongxin Yu12Ajay Jain13Qinggong Tang14https://orcid.org/0000-0001-9499-5384Chongle Pan15https://orcid.org/0000-0003-2860-0334School of Computer Science, Gallogly College of Engineering, The University of Oklahoma, Norman, OK, USAStephenson School of Biomedical Engineering, The University of Oklahoma, Norman, OK, USASchool of Computer Science, Gallogly College of Engineering, The University of Oklahoma, Norman, OK, USASchool of Computer Science, Gallogly College of Engineering, The University of Oklahoma, Norman, OK, USAStephenson School of Biomedical Engineering, The University of Oklahoma, Norman, OK, USASchool of Computer Science, Gallogly College of Engineering, The University of Oklahoma, Norman, OK, USASchool of Computer Science, Gallogly College of Engineering, The University of Oklahoma, Norman, OK, USAStephenson School of Biomedical Engineering, The University of Oklahoma, Norman, OK, USAStephenson School of Biomedical Engineering, The University of Oklahoma, Norman, OK, USAStephenson School of Biomedical Engineering, The University of Oklahoma, Norman, OK, USAStephenson School of Biomedical Engineering, The University of Oklahoma, Norman, OK, USADepartment of Pathology, University of Oklahoma Health Sciences Center, Oklahoma City, OK, USADepartment of Pathology, University of Oklahoma Health Sciences Center, Oklahoma City, OK, USAStephenson Cancer Center, University of Oklahoma Health Sciences Center, Oklahoma City, OK, USAStephenson School of Biomedical Engineering, The University of Oklahoma, Norman, OK, USASchool of Computer Science, Gallogly College of Engineering, The University of Oklahoma, Norman, OK, USAOptical coherence tomography (OCT) imaging enables high resolution visualization of sub-surface tissue microstructures. However, OCT image analysis using deep learning is hampered by limited diverse training data to meet performance requirements and high inference latency for real-time applications. To address these challenges, we developed Octascope, a lightweight domain-specific convolutional neural network (CNN) - based model designed for OCT image analysis. Octascope was pre-trained using a curriculum learning approach, which involves sequential training, first on natural images (ImageNet), then on OCT images from retinal, abdominal, and renal tissues, to progressively acquire transferable knowledge. This multi-domain pre-training enables Octascope to generalize across varied tissue types. In two downstream tasks, Octascope demonstrated notable improvements in predictive accuracy compared to alternative approaches. In the epidural tissue detection task, our method surpassed single-task learning with fine-tuning by 9.13% and OCT-specific transfer learning by 5.95% in accuracy. Octascope outperformed VGG16 and ResNet50 by 5.36% and 6.66% in a retinal diagnosis task, respectively. In comparison to a Transformer-based OCT foundation model - RETFound, Octascope delivered 2 to 4.4 times faster inference speed with slightly better predictive accuracies in both downstream tasks. Octascope represented a significant advancement for OCT image analysis by providing an effective balance between computational efficiency and diagnostic accuracy for real-time clinical applications.https://ieeexplore.ieee.org/document/11113249/Deep learningdomain-specificfoundation modellightweightOctascopeOCT medical imaging |
| spellingShingle | Haoyang Cui Chen Wang Paul Calle Yunlong Liu Qinghao Zhang Sinaro Ly Justin Reynolds Feng Yan Ke Zhang Ronghao Liu Junyuan Liu Kar-Ming Fung Zhongxin Yu Ajay Jain Qinggong Tang Chongle Pan Octascope: A Lightweight Pre-Trained Model for Optical Coherence Tomography IEEE Access Deep learning domain-specific foundation model lightweight Octascope OCT medical imaging |
| title | Octascope: A Lightweight Pre-Trained Model for Optical Coherence Tomography |
| title_full | Octascope: A Lightweight Pre-Trained Model for Optical Coherence Tomography |
| title_fullStr | Octascope: A Lightweight Pre-Trained Model for Optical Coherence Tomography |
| title_full_unstemmed | Octascope: A Lightweight Pre-Trained Model for Optical Coherence Tomography |
| title_short | Octascope: A Lightweight Pre-Trained Model for Optical Coherence Tomography |
| title_sort | octascope a lightweight pre trained model for optical coherence tomography |
| topic | Deep learning domain-specific foundation model lightweight Octascope OCT medical imaging |
| url | https://ieeexplore.ieee.org/document/11113249/ |
| work_keys_str_mv | AT haoyangcui octascopealightweightpretrainedmodelforopticalcoherencetomography AT chenwang octascopealightweightpretrainedmodelforopticalcoherencetomography AT paulcalle octascopealightweightpretrainedmodelforopticalcoherencetomography AT yunlongliu octascopealightweightpretrainedmodelforopticalcoherencetomography AT qinghaozhang octascopealightweightpretrainedmodelforopticalcoherencetomography AT sinaroly octascopealightweightpretrainedmodelforopticalcoherencetomography AT justinreynolds octascopealightweightpretrainedmodelforopticalcoherencetomography AT fengyan octascopealightweightpretrainedmodelforopticalcoherencetomography AT kezhang octascopealightweightpretrainedmodelforopticalcoherencetomography AT ronghaoliu octascopealightweightpretrainedmodelforopticalcoherencetomography AT junyuanliu octascopealightweightpretrainedmodelforopticalcoherencetomography AT karmingfung octascopealightweightpretrainedmodelforopticalcoherencetomography AT zhongxinyu octascopealightweightpretrainedmodelforopticalcoherencetomography AT ajayjain octascopealightweightpretrainedmodelforopticalcoherencetomography AT qinggongtang octascopealightweightpretrainedmodelforopticalcoherencetomography AT chonglepan octascopealightweightpretrainedmodelforopticalcoherencetomography |