Deep learning differentiates between healthy and diabetic mouse ears from optical coherence tomography angiography images.

We trained a deep learning algorithm to use skin optical coherence tomography (OCT) angiograms to differentiate between healthy and type 2 diabetic mice. OCT angiograms were acquired with a custom-built OCT system based on an akinetic swept laser at 1322 nm with a lateral resolution of ∼13 µm and using split-spectrum amplitude decorrelation. Our data set consisted of 24 stitched angiograms of the full ear, with a size of approximately 8.2 × 8.2 mm, evenly distributed between healthy and diabetic mice. The deep learning classification algorithm uses the ResNet v2 convolutional neural network architecture and was trained on small patches extracted from the full ear angiograms. For individual patches, we obtained a cross-validated accuracy of 0.925 and an area under the receiver operating characteristic curve (ROC AUC) of 0.974. Averaging over multiple patches extracted from each ear resulted in the correct classification of all 24 ears.

Hyperspectral optical coherence tomography for in vivo visualization of melanin in the retinal pigment epithelium.

Previous studies for melanin visualization in the retinal pigment epithelium (RPE) have exploited either its absorption properties (using photoacoustic tomography or photothermal optical coherence tomography [OCT]) or its depolarization properties (using polarization sensitive OCT). However, these methods are only suitable when the melanin concentration is sufficiently high. In this work, we present the concept of hyperspectral OCT for melanin visualization in the RPE when the concentration is low. Based on white light OCT, a hyperspectral stack of 27 wavelengths (440-700 nm) was created in post-processing for each depth-resolved image. Owing to the size and shape of the melanin granules in the RPE, the variations in backscattering coefficient as a function of wavelength could be identified-a result which is to be expected from Mie theory. This effect was successfully identified both in eumelanin-containing phantoms and in vivo in the low-concentration Brown Norway rat RPE.

Automated segmentation of dermal fillers in OCT images of mice using convolutional neural networks.

We present a system for automatic determination of the intradermal volume of hydrogels based on optical coherence tomography (OCT) and deep learning. Volumetric image data was acquired using a custom-built OCT prototype that employs an akinetic swept laser at ~1310 nm with a bandwidth of 87 nm, providing an axial resolution of ~6.5 µm in tissue. Three-dimensional data sets of a 10 mm × 10 mm skin patch comprising the intradermal filler and the surrounding tissue were acquired. A convolutional neural network using a u-net-like architecture was trained from slices of 100 OCT volume data sets where the dermal filler volume was manually annotated. Using six-fold cross-validation, a mean accuracy of 0.9938 and a Jaccard similarity coefficient of 0.879 were achieved.

Characterization of dry eye disease in a mouse model by optical coherence tomography and fluorescein staining.

A custom-built ultrahigh-resolution optical coherence tomography (UHR-OCT) system and fluorescein staining were employed for investigation of a scopolamine induced dry eye mouse model. Acquired data was used to evaluate common and complementary findings of the two modalities. Central corneal thickness as measured by UHR-OCT increased significantly over the study period of 24 hours, from 89.0 ± 3.57 µm to 92.2 ± 4.07 µm. Both techniques were able to show corneal lesions with a large range of severity. Localized fluorescein staining was detected in 5% and diffuse staining in 45% of cases where no epithelial damage was visible with OCT. However, OCT revealed stromal defects in 6% and endothelial defects in 18% of the cases, which could not be visualized via fluorescein staining. Thus, while fluorescein staining widely detected defects of the corneal surface in a mouse model of dry eye disease, OCT non-invasively revealed additional information about defect depth and involvement of particular layers.