Our NEO segmentation algorithm is applied to Optical Coherence Tomography (OCT) b-scans acquired with the Heidelberg Spectralis.
It was trained on our dense segmentation dataset, consisting of manually segmented full volume OCT scans from both TopCon 3D-2000 and Heidelberg Spectralis OCTs. Entire OCT volumes where manually double graded by expert graders at the Moorfields Reading Centre (all 128 b-scans of each TopCon OCT and all 49 b-scans of each Heidelberg OCT), to generate the training data for the NEO model.
It is able to segment and quantify the following features of neo-vascular AMD:
- Pigment epithelium detachment (PED)
- Subretinal hyper-reflective material (SHRM)
- Subretinal fluid (SRF)
- Intra retinal fluid (IRF)
And the following features of Macular Oedema (Diabetic, secondary to Retinal Vein Occlusion, other causes):
Deep learning model implemented in PyTorch trained on data of Heidelberg scans and validated in an independent, external, real-life dataset demonstrating human-level performance. The model runs on AWS lambda. It takes as input a macular OCT volume and produces segmentation maps.
A PDF report is returned which contains B-scans visualisation of the segmentation output.
Privacy & Terms
By using NEO you agree to the following: