Unlock instant, AI-driven research and patent intelligence for your innovation.

A network method for training fundus lesion point segmentation based on loss function

A loss function and network technology, applied in the field of neural networks, can solve the problems of many misclassifications and low learning efficiency in the segmentation network, and achieve the effect of alleviating the problem of misidentification and speeding up the learning rate

Active Publication Date: 2021-03-19
NANKAI UNIV
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] Aiming at the technical problems existing in the prior art, the present invention provides a method for training fundus lesion point segmentation network based on loss function, which can solve the problems of many misclassifications of segmentation network and low learning efficiency caused by class balance cross-entropy loss function , for efficient segmentation of fundus lesions

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A network method for training fundus lesion point segmentation based on loss function
  • A network method for training fundus lesion point segmentation based on loss function
  • A network method for training fundus lesion point segmentation based on loss function

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0052] Related techniques, such as the class-balanced cross-entropy loss function, assign high weights to a small proportion of pixels and low weights to a high proportion of pixels. However, this method does not consider the weights between samples, all negative samples are treated equally and share the same weight. Therefore, in class-balanced cross-entropy loss, negative samples are often misclassified as positive samples. In addition, the training efficiency of this loss function is not high. Therefore, in order to solve this technical problem in this embodiment, a new loss function is proposed to train the lesion point segmentation network, so that the network can focus on the learning of difficult samples, and improve the learning efficiency and anti-interference of the network.

[0053] Before introducing the technical solution of the embodiment of the present invention, it is first necessary to define the lesion point segmentation network used in the embodiment of the...

Embodiment 2

[0087] On the bleeding point segmentation task of the fundus image, use the loss function to train figure 1 The segmentation network in , the specific method includes the following steps:

[0088] Step 1: Preprocess the IDRiD fundus dataset, 54 fundus images are used as the training set, and 27 images are used as the test set. Since the size of the images is too large to be handled by computer hardware, the resolution of each image is down-sampled to 1440×960. Data augmentation is performed on the training set by means of rotation, mirroring, etc. Set the hyperparameters of the segmentation network;

[0089] Step 2: Initialize the weights of each layer of the segmentation network;

[0090] Step 3: Randomly select a fundus image from the expanded training set. And randomly crop out an 800×800 region from the image;

[0091] Step 4: The fundus image passes through the processing module in the segmentation network to obtain the input of the loss function layer, that is, the ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a loss function-based fundus lesion point segmentation network training method, which uses a loss function to train a depth segmentation network to efficiently segment fundus lesion points to judge whether the negative sample is reserved or discarded according to the result of the indicator function, if the value of the indicator function is 1, the negative sample is reserved, and otherwise, the negative sample is discarded. Therefore, the discrimination capability and the learning rate of the network are improved, easy samples are discarded at a high probability, and hard samples are discarded at a low probability; under the condition that the hard samples are reserved, a large amount of sample selection time can be saved, and therefore the network is concentratedon learning of the hard samples. According to the method, the problems of many segmentation network wrong segmentation conditions and low learning efficiency caused by the class balance cross entropyloss function can be solved, and the fundus lesion point is efficiently segmented.

Description

technical field [0001] The invention belongs to the technical field of neural networks, in particular to a method for training a fundus lesion point segmentation network based on a loss function. Background technique [0002] As a deep learning model, deep convolutional neural networks have achieved state-of-the-art performance on many computer vision tasks such as image classification, object detection, and object segmentation. In recent years, deep learning-based semantic segmentation models have been extensively studied and achieved remarkable results. However, to the best of our knowledge, most existing models focus on normal-sized objects such as animals and vehicles. Semantic segmentation for small objects has not been fully studied. For example, the detection of fundus lesions in the medical field. However, segmenting small lesions is different from segmenting normal-sized objects. There is always a class imbalance problem in small object segmentation, which is co...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06T7/00G06T7/11
CPCG06T7/0012G06T2207/20081G06T2207/20084G06T2207/20132G06T2207/30041G06T2207/30096G06T7/11
Inventor 郭松李涛王恺康宏
Owner NANKAI UNIV