Establishing method of retinal layer auto-segmentation model, retinal layer quantification system, eye care device, method for detecting retinal layer thickness and retinal layer area, and method for assessing and predicting neurodegenerative disease

An establishing method of a retinal layer auto-segmentation model, comprising:

• obtaining a reference database, wherein the reference database comprises a plurality of reference optical coherence tomographic images;
• performing a reference image pre-processing step to duplicate each of the plurality of reference optical coherence tomographic images, mark a cell segmentation line of each of retinal layers and crop each of the plurality of reference optical coherence tomographic images, so as to obtain a plurality of control label images;
• performing an image feature selecting step to analyze each of the plurality of control label images by a feature selecting module, and obtain an each layer control label image feature from each of the plurality of control label images, so as to obtain a plurality of each layer control label image features;
• performing a data set generating step to process the reference optical coherence tomographic images and corresponding one of the control label images in a data enhancement method to obtain a data set, and divide the data set into a training set and a validation set, wherein the data set comprises the plurality of reference optical coherence tomographic images, the plurality of control label images, a plurality of adjusted reference optical coherence tomographic images and a plurality of adjusted control label images;
• performing a training step to train the training set with the plurality of each layer control label image features through a U-net convolution neural network learning classifier to reach convergence, so as to obtain the retinal layer auto-segmentation model;
• and
• performing a confirming step to output a plurality of label reference images from the validation set using the retinal layer auto-segmentation model, and to compare each of the plurality of label reference images with corresponding one of the plurality of control label images, so as to confirm an accuracy of the retinal layer auto-segmentation model.