Layer boundary evolution for macular OCT segmentation

Unmet Need
Optical coherence tomography (OCT) is a three-dimensional (3D) imaging technique that emits a beam of light into the tissues to be examined and detects the reflected or back-scattered light from the tissues. OCT produces high-resolution depth images of the retina and is now the standard of care for in vivo ophthalmological assessment. It is also increasingly being used for evaluation of neurological disorders such as multiple sclerosis (MS). The images obtained via OCT are used to segment layers of maculae (area responsible for high-acuity color vision) in human eyes. Automatic segmentation methods can identify the retinal layers of the macular cube providing consistent results without intra- and inter-rater variation and is faster than manual segmentation, allowing faster and better evaluations- though these technologies are still being developed. There is a need for a method of segmenting the images obtained via OCT in an efficient and accurate manner, particularly for evaluation of multiple sclerosis.
 
Technology Overview
This invention is a fast, multilayer macular OCT algorithmic segmentation method based on a fast level set method. The framework for this method uses contours in an optimized approach specifically for OCT layer segmentation over the whole macular cube. Evaluation on both healthy and multiple sclerosis subjects shows that this method is statistically better than a state-of-the-art graph-based method.

Stage of Development
The method has been utilized to evaluate data from the right eyes of 36 patients (21 MS patients and 15 healthy controls). The method was trained using data from 9 MS patients and 6 healthy controls, using the remaining patient data as part of the validation set. A cumulative 12,960 experiments were run using this data to confirm the model. The algorithm demonstrated excellent performance in terms of both mean absolute boundary error and layer Dice coefficient. In addition, less computation cost is required, which makes it an ideal alternative for graph cut search in boundary refinement after feature classification. Although this technique does not outperform deep learning based methods in terms of speed (since the latter needs only a few seconds to run on a GPU), this method provides topologically correct results and subvoxel resolution that most deep learning based methods cannot achieve.
 
Publications
Liu Y et al. Biomed Opt Express. 2019 Mar 1;10(3): 1064–1080.

Category(s):