High myopia detection method based on eye fundus image and related device

Abstract

The invention provides a high myopia detection method based on an eye fundus image. The method comprises the following steps: acquiring a to-be-detected eye fundus image; inputting the fundus image to be detected into a high myopia detection model to obtain a high myopia detection result of the fundus image to be detected; wherein the high myopia detection model is obtained by training by using an adaptive course learning module and a semi-supervised transfer learning framework based on a general data set, a labeled fundus image sample data set and a label-free fundus image sample data set; wherein the adaptive course learning module is used for screening eye fundus image samples with labels and sequentially transmitting the screened eye fundus image samples with the labels to the student classification network model according to a determined sequence; the semi-supervised transfer learning framework comprises a pre-training source model and a semi-supervised teacher network model.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for detecting high myopia based on a fundus image and a related device. Background technique [0002] Detecting various diseases from fundus images has become a medical technology. Eyes with a diopter of 600 degrees or more are defined as high myopia. To obtain a high myopia risk prediction model that can detect the degree of myopia from fundus images, the risk When predicting the model, most of the existing technologies directly put the acquired data set into the current popular deep learning model for training, but in the real world, the fundus image data set of high myopia is very small, and in the clinically collected data set, The non-high myopia fundus image data set and the high myopia fundus image data set are extremely unbalanced. Directly using a large classification network to train the high myopia risk prediction model has serious overfitting problem...

AI Technical Summary

Problems solved by technology

[0003] The present invention provides a high myopia detection method based on fundus images and a related device, which are used to solve the problem of serious overfitting in the high myopia risk prediction model directly trained by a large classification network in the prior art, and the difficulty in distinguishing hard sample defect

Images