System for detecting whether coronary angiography has complete occlusion lesion or not based on deep learning

Abstract

The invention discloses a system for detecting whether coronary angiography has a complete occlusion lesion or not based on deep learning. A deep learning recurrent neural network is used for analyzing an overall video, a GPU (graphics processing unit) is used for accelerating calculation to obtain a detection result, the calculation delay is small, and the real-time problem of detection is solved. The system comprises a video input module, a convolutional neural network module, a coding and decoding attention module and a classification module. Particularly, the encoding and decoding attention module and the supervision algorithm thereof provided by the invention are used for generating a proper attention weight to improve the detection accuracy. The end-to-end system design achieves automation of the coronary artery complete occlusion detection process, and complex intermediate steps are not needed. Compared with other systems for detecting whether the complete occlusion lesion exists or not, the whole video instead of a single image is analyzed, the situation of misjudgment caused by the single image is reduced, and the accuracy of existence judgment of the complete occlusion lesion is remarkably improved.

Description

technical field [0001] The invention relates to the technical field of digital image processing, in particular to a system for detecting whether there is a total occlusion lesion in coronary angiography based on deep learning. Background technique [0002] Coronary angiography videos are commonly used medical data for diagnosing cardiovascular disease. Cardiovascular lesions such as calcification, thrombus, stenosis and complete occlusion can all be observed through coronary angiography video. Thanks to the development of artificial intelligence and machine learning, there are already many applications that use deep learning methods to analyze medical data and assist in diagnosis. [0003] Existing applications are generally based on medical images, and use convolutional neural networks to extract abstract features and process them for classification and detection. Due to the technology and equipment of medical data, the image taken has a lot of noise, and it is easy to ca...

AI Technical Summary

Problems solved by technology

However, in the coronary angiography video, we only need to focus on the key video frames in the filling stage, and the analysis of the entire video increases the amount of calculation.

In addition, for completely occluded lesions, the observed morphology of the lesion is reflected by the absence of contrast agent and incomplete vascular morphology, which is similar to the vascular morphology in the diffusion and dissipation stages, resulting in misjudgment, so it is flawed to analyze the entire video

Images