This study introduces a learning-assisted denoising technique for skin Optical Coherence Tomography (OCT) images. By combining Reinforcement Learning (RL) with the Denoising Convolutional Neural Network (DnCNN), we achieve enhanced denoising capabilities. The method iteratively refines DnCNN parameters through RL-guided policies, demonstrating superior performance. Tailored for skin OCT images, the approach prioritizes preserving vital structures for accurate clinical assessments. This integration of RL into DnCNN training represents a promising advancement in medical image denoising, particularly for dermatological diagnostics.
The position and appearance of the dermal epidermal junction (DEJ) is an important indicator of skin health. Optical coherence tomography (OCT) is used for noninvasive skin imaging but is impeded by the training required. DEJ delineation algorithms address this issue but with limited consideration of diversity in samples. In this study, marked images of a variety of body regions, age groups, and Fitzpatrick skin type (FST) groups were used. To find the DEJ automatically, image columns were matched to a body region-specific swatch of similar columns based on cosine similarity. Our results demonstrated the swatches method could determine around 87% of all automatic markings within 39 micrometers axially of the manually identified DEJ.
KEYWORDS: Optical coherence tomography, Skin, Deep learning, Education and training, Speckle, Tumor growth modeling, Image quality, Signal to noise ratio, Image processing, Data modeling
Optical coherence tomography (OCT) is well-known for its high-resolution, non-invasive imaging modality with many medical uses, including skin imaging. Nevertheless, speckle noise limits the analytical capabilities of this imaging tool, causing deterioration in contrast and less exact detection of tissue microstructural heterogeneity. To address this issue, we proposed OCT despeckling approach by combing it with normalization to reduce the speckle noise more effectively. The proposed method contains multiple steps including phase correlation for alignment of misaligned frames, frame averaging which minimizes speckle noise, region-wise pixels normalization that helps to normalize intensity pixels, a modified BM3D filtering to suppress the white and speckle, and contrast enhancement to improve the contrast appropriately. To establish the approach, we applied 130 distinct B-scan skin OCT images and validate and evaluate the performance using qualitatively and quantitatively. Although the output obtained by the algorithm is promising, the method is time-consuming because of a series of steps. To reduce the time complexity, we also develop a supervised deep learning model by mapping between noisy-despeckled image pairs. The effectiveness and applicability of our DL approach were assessed using 130 skin OCT B-scans from various body areas taken from 45 healthy people between the ages of 20 and 60. With the support of the experimental results, we demonstrate that our DL model is capable to normalize and despeckling OCT images simultaneously.
Optical coherence tomography (OCT) images enable the visualization of cell layers, and accurate layer thickness is crucial for disease diagnosis and treatment tracking. To measure layer thickness, delineating the layer boundaries is the first step. In this paper, we proposed a time-efficient layer segmentation method developed on central unit processors (CPUs). This method consists of convolutional neural networks (CNN) and graph search (CNN-GS). CNN-GS aims to automatically segment two defined boundaries to calculate the epidermal thickness. We applied our method to 110 skin OCT images from various body locations, taken from 13 healthy individuals aged between 20 and 60 years, to evaluate the performance and versatility of our method. Our method demonstrated an overall 94.68% accuracy on patch-wise classification and an 85.81% accuracy on segmentation position accuracy as compared to manual segmentation, allowing 94.87% accuracy on epidermal thickness. In addition, our method performed a near real-time image analysis, costing less than 1 second per skin OCT image to delineate the layer boundaries.
OCT is a promising imaging modality for the field of dermatology, but lack of research is hinders its use in everyday practice. Our study provides qualitative and quantitative analysis of healthy skin from various anatomic locations. Qualitative analysis showed identification of key structures including epidermis, dermis, and dermal epidermal junction (DEJ), as well as accessory structures. Quantitative analysis produced a characteristic absorption spectra of healthy skin, with a peaked intensity at stratum corneum, second peak at DEJ, and variations due to accessory structures. This analysis is valuable for OCTs use as a diagnostic aid in determining healthy vs pathologic skin.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.