Pulmonary embolism detection system based on convolutional neural network

Abstract

The invention discloses a pulmonary embolism detection system based on a convolutional neural network. The system comprises: a candidate region extraction network, which is a full convolutional network using automatic encoding and decoding with jump connection, performing candidate region extraction on a computed tomography angiography image to be detected, and obtaining a plurality of false positive candidate regions with different sizes; a 3D affine transformation network, used for generating cubes with aligned blood vessels and fixed sizes from a plurality of false positive candidate areaswith different sizes and taking out three orthogonal layers of the cubes; and a false positive prediction screening network, used for inputting the three orthogonal layers into a 2D classification network containing two full connection layers to carry out false positive prediction screening. The method can solve the problem of error accumulation. 3D image features with higher discriminability canbe automatically extracted, the influence of the volume effect is reduced, and the method does not depend on the experience of researchers. The accuracy is improved while the recall rate is ensured.

Description

technical field [0001] The invention belongs to the field of computer technology, and more specifically relates to a pulmonary embolism detection system based on a convolutional neural network. Background technique [0002] Computed Tomography Pulmonary Arteriography (CTPA) is the primary means of diagnosing pulmonary embolism in clinical practice today. However, for suspected pulmonary embolism, tracing the pulmonary artery in 300-500 slices is very time-consuming, and the diagnostic result will be affected by the radiologist's experience, attention span, and fatigue, etc. Therefore, in clinical practice, in order to improve the diagnostic accuracy and efficiency of pulmonary embolism, some automatic detection methods have been proposed. Existing methods usually include two separate steps: (1) generating a set of PE candidate regions based on voxel-level features); (2) extracting region-level features of suspected points and eliminating false positive predictions based on ...

AI Technical Summary

Problems solved by technology

Although some effects have been achieved, the system still has the following problems: the first step for generating candidate regions still uses manually designed features; the two steps of pulmonary embolism detection are independent of each other, and the performance of the other step is not considered. The effect of the step; in order to achieve an acceptable recall, it produces a large number of false positive results

Images