Semi-supervised Assisted Multi-Task Learning for Oral Optical Coherence Tomography Image Segmentation and Denoising DOI Creative Commons
Jinpeng Liao, Tianyu Zhang, Simon Shepherd

и другие.

Biomedical Optics Express, Год журнала: 2024, Номер 16(3), С. 1197 - 1197

Опубликована: Дек. 5, 2024

Optical coherence tomography (OCT) is promising to become an essential imaging tool for non-invasive oral mucosal tissue assessment, but it faces challenges like speckle noise and motion artifacts. In addition, difficult distinguish different layers of tissues from gray level OCT images due the similarity optical properties between layers. We introduce Efficient Segmentation-Denoising Model (ESDM), a multi-task deep learning framework designed enhance by reducing scan time ∼8s ∼2s improving epithelium layer segmentation. ESDM integrates local feature extraction capabilities convolution long-term information processing advantages transformer, achieving better denoising segmentation performance compared existing models. Our evaluation shows that outperforms state-of-the-art models with PSNR 26.272, SSIM 0.737, mDice 0.972, mIoU 0.948. Ablation studies confirm effectiveness our design, such as fusion methods, which minimal model complexity increase. also presents high accuracy in quantifying thickness, mean absolute errors low 5 µm manual measurements. This research can notably improve reduce cost accurate epithermal segmentation, diagnostic clinical settings.

Язык: Английский

Sub2Full: split spectrum to boost optical coherence tomography despeckling without clean data DOI
Lingyun Wang, José‐Alain Sahel, Shaohua Pi

и другие.

Optics Letters, Год журнала: 2024, Номер 49(11), С. 3062 - 3062

Опубликована: Май 2, 2024

Optical coherence tomography (OCT) suffers from speckle noise, causing the deterioration of image quality, especially in high-resolution modalities such as visible light OCT (vis-OCT). Here, we proposed an innovative self-supervised strategy called Sub2Full (S2F) for despeckling without clean data. This approach works by acquiring two repeated B-scans, splitting spectrum first repeat a low-resolution input, and utilizing full second target. The method was validated on vis-OCT retinal images visualizing sublaminar structures outer retina demonstrated superior performance over state-of-the-art Noise2Noise (N2N) Noise2Void (N2V) schemes.

Язык: Английский

Процитировано

4
Self-supervised denoising with Edge Perception in OCT images DOI
Feiyi Xu, Zhaofei Wu, Shuai You

и другие.

Computers & Electrical Engineering, Год журнала: 2025, Номер 124, С. 110360 - 110360

Опубликована: Май 1, 2025

Язык: Английский

Процитировано

0
BreakNet: Discontinuity-Resilient Multi-Scale Transformer Segmentation of Retinal Layers DOI Creative Commons
Razieh Ganjee, Bingjie Wang, Lingyun Wang

и другие.

Biomedical Optics Express, Год журнала: 2024, Номер 15(12), С. 6725 - 6725

Опубликована: Окт. 30, 2024

Visible light optical coherence tomography (vis-OCT) is gaining traction for retinal imaging due to its high resolution and functional capabilities. However, the significant absorption of hemoglobin in visible range leads pronounced shadow artifacts from blood vessels, posing challenges accurate layer segmentation. In this study, we present BreakNet, a multi-scale Transformer-based segmentation model designed address boundary discontinuities caused by these artifacts. BreakNet utilizes hierarchical Transformer convolutional blocks extract global local feature maps, capturing essential contextual, textural, edge characteristics. The incorporates decoder that expand pathways enhance extraction fine details semantic information, ensuring precise Evaluated on rodent images acquired with prototype vis-OCT, demonstrated superior performance over state-of-the-art models, such as TCCT-BP U-Net, even when faced limited-quality ground truth data. Our findings indicate has potential significantly improve quantification analysis.

Язык: Английский

Процитировано

1
Semi-supervised Assisted Multi-Task Learning for Oral Optical Coherence Tomography Image Segmentation and Denoising DOI Creative Commons
Jinpeng Liao, Tianyu Zhang, Simon Shepherd

и другие.

Biomedical Optics Express, Год журнала: 2024, Номер 16(3), С. 1197 - 1197

Опубликована: Дек. 5, 2024

Optical coherence tomography (OCT) is promising to become an essential imaging tool for non-invasive oral mucosal tissue assessment, but it faces challenges like speckle noise and motion artifacts. In addition, difficult distinguish different layers of tissues from gray level OCT images due the similarity optical properties between layers. We introduce Efficient Segmentation-Denoising Model (ESDM), a multi-task deep learning framework designed enhance by reducing scan time ∼8s ∼2s improving epithelium layer segmentation. ESDM integrates local feature extraction capabilities convolution long-term information processing advantages transformer, achieving better denoising segmentation performance compared existing models. Our evaluation shows that outperforms state-of-the-art models with PSNR 26.272, SSIM 0.737, mDice 0.972, mIoU 0.948. Ablation studies confirm effectiveness our design, such as fusion methods, which minimal model complexity increase. also presents high accuracy in quantifying thickness, mean absolute errors low 5 µm manual measurements. This research can notably improve reduce cost accurate epithermal segmentation, diagnostic clinical settings.

Язык: Английский

Процитировано

0