Ultrasonic image deep learning distributed training system and training method

Abstract

The invention provides an ultrasonic image deep learning distributed training system. The ultrasonic image deep learning distributed training system comprises a cloud coordination node and a plurality of training nodes which are connected, the cloud coordination node is at least used for storing a corresponding initial diagnosis model for disease species and coordinating the training task of each training node, comprising: issuing an initial diagnosis model of a corresponding disease category and corresponding training configuration information to each training node according to the state of the training node, recovering a local diagnosis model obtained by training each training node, and updating a self-stored diagnosis model of the corresponding disease category; and the training node is at least used for training by using the classified ultrasonic image data marked by the local corresponding disease category according to the received initial diagnosis model structure and the corresponding training configuration information to obtain a corresponding local diagnosis model, form model updating information and upload the model updating information to the cloud coordination node.

Description

technical field [0001] The invention relates to the technical field of ultrasonic medical imaging, in particular to a distributed training system and training method for deep learning of ultrasonic images. Background technique [0002] With the continuous development and progress of ultrasonic medical imaging technology and computer technology, ultrasonic image data is being generated on an unprecedented scale. In recent years, deep learning has rapidly developed into a research hotspot in ultrasonic medical image analysis. It can learn from The hidden disease diagnosis features are automatically extracted from the big data of ultrasonic medical images, which provides good auxiliary tools for doctors' daily work. [0003] Although the generation speed of medical imaging data is getting faster and the storage capacity is getting bigger and bigger, the distribution of data is relatively scattered, and most of them are in the hands of medical institutions with good medical equi...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to overcome the deficiencies in the prior art, provide a distributed training system for deep learning of ultrasound images, and a training method, which solves the difficulties in data collection, data privacy and security, and image quality caused by traditional distributed training. data breach risk

Images