Neural network model training method and device for pathological image sample

Abstract

The invention relates to a neural network model training method and device for pathological image samples with label noise, and the method comprises the steps of inputting the pathological image samples with initial labels into an initial neural network model, and obtaining the initial prediction probability of each sample; calculating an initial cross entropy loss function and an initial gradientmodule length according to the initial label and the initial prediction probability of each sample; calculating the gradient density corresponding to each sample by using the gradient module length,carrying out weighted calculation on the initial cross entropy loss function by using the gradient density and Gaussian probability distribution, and adjusting the influence of the easy-to-separate sample and the extremely difficult-to-separate sample on model training; optimizing the initial label of the sample by using the gradient die length; training the model by using a gradient descent method and the weighted corrected cross entropy loss function, and updating parameters of the model; in the next iteration, taking the optimized label as the label of the sample to solve the loss value, repeating the iterative training for several times until the model converges, thus improving the model training effect.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a neural network model training method and device for pathological image samples. Background technique [0002] Artificial intelligence-assisted pathological image analysis is one of the more popular research fields at present. Relying on artificial intelligence assistance can save a lot of time for pathologists to read images, improve the work efficiency of pathologists, and reduce the errors caused by the subjective interpretation of different doctors. At present, most of the mainstream artificial intelligence-assisted pathological image analysis is based on the supervised learning method of deep learning. Supervised learning first needs to manually label a large number of training samples, and then use these labeled samples to train the model. After the model training is completed, the pathological images to be analyzed can be input into the model for analysis. [00...

AI Technical Summary

Problems solved by technology

Forcing the model to learn these samples with label noise will not only learn some wrong labels, but also affect the accuracy of the model on the data that can already be correctly classified

Images