Real-time retinal layer segmentation of OCT volumes with GPU accelerated inferencing using a compressed, low-latency neural network.

Segmentation of retinal layers in optical coherence tomography (OCT) is an essential step in OCT image analysis for screening, diagnosis, and assessment of retinal disease progression. Real-time segmentation together with high-speed OCT volume acquisition allows rendering of en face OCT of arbitrary retinal layers, which can be used to increase the yield rate of high-quality scans, provide real-time feedback during image-guided surgeries, and compensate aberrations in adaptive optics (AO) OCT without using wavefront sensors. We demonstrate here unprecedented real-time OCT segmentation of eight retinal layer boundaries achieved by 3 levels of optimization: 1) a modified, low complexity, neural network structure, 2) an innovative scheme of neural network compression with TensorRT, and 3) specialized GPU hardware to accelerate computation. Inferencing with the compressed network U-NetRT took 3.5 ms, improving by 21 times the speed of conventional U-Net inference without reducing the accuracy. The latency of the entire pipeline from data acquisition to inferencing was only 41 ms, enabled by parallelized batch processing. The system and method allow real-time updating of en face OCT and OCTA visualizations of arbitrary retinal layers and plexuses in continuous mode scanning. To the best our knowledge, our work is the first demonstration of an ophthalmic imager with embedded artificial intelligence (AI) providing real-time feedback.

Pseudo-real-time retinal layer segmentation for high-resolution adaptive optics optical coherence tomography.

We present a pseudo-real-time retinal layer segmentation for high-resolution Sensorless Adaptive Optics-Optical Coherence Tomography (SAO-OCT). Our pseudo-real-time segmentation method is based on Dijkstra's algorithm that uses the intensity of pixels and the vertical gradient of the image to find the minimum cost in a geometric graph formulation within a limited search region. It segments six retinal layer boundaries in an iterative process according to their order of prominence. The segmentation time is strongly correlated to the number of retinal layers to be segmented. Our program permits en face images to be extracted during data acquisition to guide the depth specific focus control and depth dependent aberration correction for high-resolution SAO-OCT systems. The average processing times for our entire pipeline for segmenting six layers in a retinal B-scan of 496 × 400 and 240 × 400 pixels are around 25.60 and 13.76 ms, respectively. When reducing the number of layers segmented to only two layers, the time required for a 240 × 400 pixel image is 8.26 ms.