This repository consists of code to segment 3 retinal layers (and the background, and outer borders), i.e. ILM-OPL-HFL, ONL, and BMEIS-OB-RPE. To see how to train the model, refer to the notebooks folders. Further, this repository consists of code to reproduce the results of the paper "Deep Learning based retinal layer segmentation in optical coherence tomography scans of patients with inherited retinal diseases" which was published and accepted in "Klinische Monatsblätter für Augenheilkunde".

https://www.thieme-connect.de/products/ejournals/pdf/10.1055/a-2227-3742.pdf

We used 2D slides of OCT scans as Input from healthy patients with the eventual goal of segmenting the layers of IRD (inherited retinal diseases) patients. Finally, the repository contains code to calculate Thickness maps and ETDRS grid regions and average thicknesses.

You can also run the code within these Colab notebooks. Here, we have included links to two Notebooks, which depict our final model setup. Please adjust the Runtime type to GPU to have a faster computation time.

• https://colab.research.google.com/drive/12wnQltPpyxaw5SCx9InbtBulSAwEfPdO Notebook to train the initial model on healthy individuals
• https://colab.research.google.com/drive/1ujpqwPge-6quofOe8VgFs4Ci9GNspajt Notebook to train the initial model on diseased individuals without using prior weights or information
• https://colab.research.google.com/drive/1gQkKExlk2yxnA8xzY03C4VMcjVNizmbj#scrollTo=j-nO5ZBTgD4U Notebook to train the initial model on diseased individuals by reusing prior weights of healthy model and freezing all layers except Batchnorm-layers
• https://colab.research.google.com/drive/1jz9xNO-1lm5vLg3vLqVNCZgqLv3mYTbs Notebook to load the trained models and segment images

(a) Exemplary slide in the middle of an OCT scan in a healthy patient with the ONL-’’hill’’ (b) Exemplary first slide of an OCT scan without ONL-’’hill’’. The same color codes for retinal layers will be used within this work. OCT = optical coherence tomography; ONL = outer nuclear layer Slide containing the ‘’ONL’’ hill (a) Prediction of the model. (b) Ground truth. (c) Input slide. ONL = outer nuclear layer Regular slide. (a) Prediction of the model. (b) Ground truth. (c) Input slide. Trained model on the IRD dataset. (a) Prediction of the model. (b) Ground truth. (c) Input slide. IRD = inherited retinal disease Trained model on the IRD dataset where ONL hill is visible. (a) Prediction of the model. (b) Ground truth. (c) Input slide. IRD = inherited retinal disease; ONL = outer nuclear layer Trained model on a IRD dataset where ONL layer is almost vanished. (a) Prediction of the model. (b) Ground truth. (c) Input slide. IRD = inherited retinal disease; ONL = outer nuclear layer Trained model on a IRD dataset: Model has problems when quality is too low. White frame indicates some wrongly segmented pixels. (a) Prediction of the model. (b) Ground This article is protected by copyright. All rights reserved. Accepted Manuscript truth. (c) Input slide. IRD = inherited retinal disease ETDRS regions. Thickness maps based on predicted segmentation of all retinal layers. (a) Thickness map of a healthy individual. (b) Thickness map of an IRD patient. Interestingly, the diseased individual has a similar full-retinal thickness. Healthy: C0 Average thickness: 274µm; S2 Average thickness 301µm; S1 Average thickness 354µm; N1 Average thickness: 352µm; N2 Average thickness: 325µm; I1 Average thickness: 352µm; I2 Average thickness: 302µm; T1 Average thickness: 348µm; T2 Average thickness: 294µm Diseased: C0 Average thickness: 281µm; S2 Average thickness 270µm; S1 Average thickness 359µm; N1 Average thickness: 369µm; N2 Average thickness: 310µm; I1 Average thickness: 366µm; I2 Average thickness: 279µm; T1 Average thickness: 343µm; T2 Average thickness: 278 IRD = inherited retinal disease Thickness maps based on predicted segmentation of ONL layer. (a) Thickness map of a healthy individual. (b) Thickness map of an IRD patient. Evidently, the IRD patient has a thinner ONL thickness, especially at the non-central regions. This article is protected by copyright. All rights reserved. Accepted Manuscript Healthy: C0 Average thickness: 109µm; S2 Average thickness 80µm; S1 Average thickness 88µm; N1 Average thickness: 94µm; N2 Average thickness: 75µm; I1 Average thickness: 88µm; I2 Average thickness: 68µm; T1 Average thickness: 89µm; T2 Average thickness: 76 µm Diseased: C0 Average thickness: 115µm; S2 Average thickness 37µm; S1 Average thickness 81µm; N1 Average thickness: 94µm; N2 Average thickness: 48µm; I1 Average thickness: 92µm; I2 Average thickness: 42µm; T1 Average thickness: 89µm; T2 Average thickness: 51 µmro IRD = inherited retinal disease

This repository contains code snippets from the following two repositories:

• https://github.com/yu02019/BEN (Model definitions)
• https://github.com/theislab/LODE/tree/master/ (Thickness map calculations)