Shift learning method and device for medical images

Abstract

The application discloses a shift learning method and device for medical images. The method comprises following steps: reading original data information of medical images; performing shift processing on the data through analyzing data attributes and integrating the data into a data format which can be accepted by a model to be analyzed; selecting a shift mode through parameter comparison on the model to be analyzed and a trained mode so as to perform parameter conversion and applying shift learning in the training of the model to be analyzed of medical images; when the model training is finished, applying the trained model parameters in the analysis of the image class. By means of the method, the accuracy of model obtained through deep learning based on a small amount of medical images can be increased. The invention also comprises a device for performing shift learning on medical images, comprising a data processing module, a shift learning module and an application module.

Description

technical field [0001] The invention relates to the fields of medical artificial intelligence and big data processing, in particular to a method and device for medical image conversion and learning. Background technique [0002] With the strong rise of the new artificial intelligence technology with the deep learning framework as the core, it has achieved considerable development and advancement in various fields. AlphaGo, driverless cars, speech recognition and other technologies that people have been looking forward to for many years are also in a very short period of time. breakthrough in time. In the foreseeable future, deep learning will also promote the development of big data analysis and artificial intelligence applications in the medical industry. [0003] However, even with the above breakthroughs, the current deep learning technology still has relatively large problems in the training process: [0004] i) The cost of model training is high. The best deep learnin...

AI Technical Summary

Problems solved by technology

[0007] 1) The medical field has more dimensions than general application scenarios. The diagnosis and treatment data of each patient are complex, and the deep learning and data analysis models are also larger and more complex than the general ones. The training cost is very high. Traditional methods are not suitable for deep learning and large-scale Learning and training data and machine learning models will consume a lot of manpower and material resources, greatly reducing the economic feasibility of the application

[0008] 2) Although the overall data volume in the medical field is large, the data sources are scattered in various hospitals, lacking a unified database available as a whole, the amount of data (such as the number of cases) around a single application scenario is very limited, and the amount of data changes is not enough to support training And computing huge deep learning and big data models, which greatly limits the scenarios where deep learning and big data technologies can be used

[0011] Modeling in the medical industry often requires the establishment of complex and huge deep learning and big data models to learn and train data due to complex scenarios. The parameter optimization link in the model training module brings a lot of pressure. Often the amount of data we have is not enough to support the samples required by the entire model to complete the optimization. The model often cannot calculate enough effects. best point

Images