To solve the phase unwrapping problem for phase images in Fourier domain Doppler optical coherence tomography (DOCT), we propose a deep learning-based residual en-decoder network (REDN) method. In our approach, we reformulate the definition for obtaining the true phase as obtaining an integer multiple of 2π at each pixel by semantic segmentation. The proposed REDN architecture can provide recognition performance with pixel-level accuracy. To address the lack of phase images that are noise and wrapping free from DOCT systems for training, we used simulated images synthesized with DOCT phase image background noise features. An evaluation study on simulated images, DOCT phase images of phantom milk flowing in a plastic tube and a mouse artery, was performed. Meanwhile, a comparison study with recently proposed deep learning-based DeepLabV3+ and PhaseNet methods for signal phase unwrapping and traditional modified networking programming (MNP) method was also performed. Both visual inspection and quantitative metrical evaluation based on accuracy, specificity, sensitivity, root-mean-square-error, total-variation, and processing time demonstrate the robustness, effectiveness and superiority of our method. The proposed REDN method will benefit accurate and fast DOCT phase image-based diagnosis and evaluation when the detected phase is wrapped and will enrich the deep learning-based image processing platform for DOCT images.

Optical coherence tomography (OCT) is a three-dimensional non-invasive high-resolution imaging modality that has been widely used for applications ranging from medical diagnosis to industrial inspection. Common OCT systems are equipped with limited field-of-view (FOV) in both the axial depth direction (a few millimeters) and lateral direction (a few centimeters), prohibiting their applications for samples with large and irregular surface profiles. Image stitching techniques exist but are often limited to at most 3 degrees-of-freedom (DOF) scanning. In this work, we propose a robotic-arm-assisted OCT system with 7 DOF for flexible large FOV 3D imaging. The system consists of a depth camera, a robotic arm and a miniature OCT probe with an integrated RGB camera. The depth camera is used to get the spatial information of targeted sample at large scale while the RGB camera is used to obtain the exact position of target to align the image probe. Eventually, the real-time 3D OCT imaging is used to resolve the relative pose of the probe to the sample and as a feedback for imaging pose optimization when necessary. Flexible probe pose manipulation is enabled by the 7 DOF robotic arm. We demonstrate a prototype system and present experimental results with flexible tens of times enlarged FOV for plastic tube, phantom human finger, and letter stamps. It is expected that robotic-arm-assisted flexible large FOV OCT imaging will benefit a wide range of biomedical, industrial and other scientific applications.

Optical coherence tomography (OCT) as an interferometric imaging technique, suffers from massive noise. Denoising methods are applied essentially to improve image quality in OCT community. The conventional methods rely on post image processing algorithms such as non-local mean filtering, block-matching and 3D filtering algorithm. However, these conventional noise reduction methods could inevitably cause the destruction of image details, reduce the contrast at the edge of OCT images, and result in a degeneration of image quality. Current deep learning methods often ignore the specificity of system, therefore haven't taken advantages of the unique characteristics of different systems. In this work, we present a deep learning noise reduction method using the network architecture trained from synthetic OCT signals with random noise that are generated from the noise formation model characterized by our custom-built specific SD-OCT (Spectrum-Domain optical coherence tomography) system. We analyze the signal formation process and the noise generation pathway of our system, thereby enabling the construction of a noise formation model. DN-Unet (Denoising Unity Network) is applied to train the datasets generated by our proposed noise formation model and the multi-to-single strategy is developed to enhance the network capability. Preliminary empirical results collectively show that the network can reach an average of 25 dB signal to noise ratio (SNR) improvement while preserving detail structures, which demonstrates the effectiveness of our noise reduction method. This method has the potential to be adopted by other systems without the need for large number of golden-standard image generation.

No abstract