A pathological image automatic recognition system and its training method based on deep learning

Abstract

The present invention proposes a training method for a pathological image automatic recognition system based on deep learning, which is characterized in that it comprises the following steps: step (a), image segmentation; step (b), obtaining a tumor region of interest and a lymph node region of interest; Step (c), extract radiomics features in the tumor region of interest and lymph node region of interest; step (d), eliminate redundant radiomics features; step (e), determine the tumor region of interest and lymph node region of interest The optimal radiomics features in the region; step (f), construct a training set with the optimal radiomics features to predict lymph node metastasis; step (g), construct an independent validation set to verify the training set; step (h), When the predictive validity of the training set reaches the preset value, the training ends; when the predictive validity of the training set is lower than the preset value, the training set is rebuilt for training. The training method of the present invention can improve the effectiveness of predicting gastric cancer lymph node metastasis.

Description

technical field [0001] The present invention relates to the field of image recognition, in particular to a deep learning-based training method for an automatic pathological image recognition system, and an automatic pathological image recognition system trained by the training method. Background technique [0002] Gastric cancer is one of the most common malignant tumors in the world, with a poor prognosis and a serious threat to human health. According to the latest statistics from GLOBOCAN, in 2018, there were about 1.033 million new cases of gastric cancer worldwide, and about 783,000 deaths, ranking fifth in terms of incidence and second in mortality among malignant tumors. According to the latest cancer statistics in my country, there are about 679,000 new cases of gastric cancer each year, and about 498,000 deaths. The number of cases and deaths ranks second among all malignant tumors. In gastric cancer, lymph node metastasis is considered to be an important prognosti...

AI Technical Summary

Problems solved by technology

[0003] Full-scan slicing technology can realize the digitization of pathological slices. With the development of full-slice image acquisition, processing and analysis technology, digital pathology has been gradually applied to scientific research and clinical practice in the past ten years. However, at present, in daily pathological slices During the diagnosis process, doctors usually detect and classify the metastatic lymph nodes in the whole slice by reading one by one, and screen out the image and location of the metastatic lymph nodes. Therefore, the workload of doctors is very heavy and the efficiency is low

Images