Small-sample meta-learning training method for artificial intelligence in medical image classification

Abstract

A small-sample meta-learning training method for medical image classification processing artificial intelligence, building three meta-learners of multi-scale CNN feature extractor, metric learner and classification discriminator, and designing the metric standard on the meta-learner; in the training set On each task, the target set is learned through the support set distance metric, and finally a metric is learned, and then for the new task of the test set, only a small number of samples in the support set can be used to quickly and correctly classify the target set; by using meta Learning to improve the practicability and reliability of artificial intelligence automatic detection in the field of intractable diseases, overcome the lack of single disease data and the lack of scattered diseases, and complete the training of medical image processing intelligent system classification with few samples, which significantly improves the accuracy of filtering. Effectively enhance the accuracy of medical image classification, greatly improve production efficiency, and contribute to the upgrading of deep learning technology for intelligent diagnosis in the medical industry.

Description

technical field [0001] The present invention relates to the structural improvement technology of equipment and devices, belonging to the G06T7 / 11 machine learning, pattern recognition and medical image processing technology in the IPC classification, or G06K9 / 00 for reading or recognizing printed or written characters or for recognizing graphics technology field, Especially the small-sample meta-learning training method for medical image classification processing artificial intelligence. Background technique [0002] At present, medical imaging including X-ray imaging, CT imaging, magnetic resonance imaging, ultrasound imaging, and nuclear medicine imaging allows doctors to understand the changes in the internal shape, function, and metabolism of the patient's body in addition to contact and anatomy. have an important role. Medical imaging plays an extremely important role in medical clinical diagnosis, and modern medicine cannot do without medical imaging technology. The ...

AI Technical Summary

Problems solved by technology

[0020] The purpose of the present invention is to provide a small-sample meta-learning training method for medical image classification and artificial intelligence, aiming at the phenomenon of relatively few data in the medical field, to solve the problem of small medical samples, to realize small-sample classification by using meta-learning methods, and to improve the accuracy of difficult diseases. The practicability and reliability of artificial intelligence automatic detection in the field can effectively enhance the accuracy of medical image classification

Images