Medical image analysis method and device

Abstract

The present invention relates to a medical image analysis method and device. The method includes the following steps that: medical images and the metadata of the content of the medical images are acquired, and the acquired medical images are matched with anatomical information images captured from a cloud medical database; edge-preserving diffusion image filtering processing is performed on medical images which are successfully matched with the anatomical information images, so that training sample medical images to be trained are obtained; network structures of a convolution layer, a de-convolution layer, a pooling layer and a whole-connection layer are sequentially adopted to perform deep training on the training sample medical images, so that a lesion point feature model is generated; and a medical image to be tested is inputted into the lesion point feature model, and a lesion point recognition result is automatically obtained. With the above method adopted, the accuracy of an algorithm can be greatly improved, false negatives can be decreased, a lesion point can be obtained accurately and rapidly, and further analysis operation can be performed on the lesion point; and a deep network is adopted, so that the problem of insufficiency of shallow-layer network feature extraction can be solved.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a medical image analysis method and device. Background technique [0002] Using algorithms to automatically identify lesions or potential lesions such as fatty liver and intestinal polyps from medical images is a problem that people have been trying to solve for many years. The traditional computer automatic recognition algorithm is to convert the original image input, that is, the pixel value into manually defined features, such as SIFT, HOG features, etc.; and then put these transformed features into a pre-trained shallow detector. Detection, the detection process is generally, slide a detection window with a preset size on the original picture, if the detection score calculated at a certain position is higher than a certain preset threshold, it is considered that this position exists. Lesion of interest or potential lesion. [0003] However, the application ...

AI Technical Summary

Problems solved by technology

[0003] However, the application of existing technologies in gastrointestinal endoscopy will produce a large number of false positives, mainly because the models used by such methods usually only contain a hidden layer for extracting features, and the extracted features are often not enough to describe and distinguish lesions point and normal area

Images