Intracranial hemorrhage subtype classification algorithm based on bilinear pooling applied to CT images

Abstract

The invention discloses a subtype classification algorithm of intracranial hemorrhage applied to CT images based on bilinear pooling. The algorithm innovatively solves the problem of subtype classification of intracranial hemorrhage from the perspective of fine-grained classification, and is used in the CNN part. The network architecture of compact bilinear pooling has been improved, the feature expression ability of neural network has been improved, and the classification performance has been improved. Bilinear pooling has been proven to be an effective method for solving fine-grained classification problems. It treats features from two CNNs from different sources or CNNs from the same source as two different features by modeling high-order statistics. And calculate the outer product of the two features, and then perform feature fusion through the pooling operation to capture the relationship between different features, and then obtain a stronger global feature representation. The method uses image translation invariance to interactively model local paired features.

Description

technical field [0001] The invention belongs to the field of intelligent medical image processing, and relates to a bilinear pooling-based classification algorithm applied to subtypes of intracranial hemorrhage. Background technique [0002] Intracranial hemorrhage (ICH) is a bleeding disorder that occurs in the ventricle or brain tissue. Intracranial hemorrhage includes: intraventricular hemorrhage (IVH), parenchymal hemorrhage (IPH), subarachnoid hemorrhage (SAH), epidural hemorrhage (EDH) and subdural hematoma (SDH). The study of subtype classification of intracranial hemorrhage is to judge whether there is intracranial hemorrhage in CT images and to classify its five subtypes (IPH, IVH, EDH, SDH, SAH). Diagnosing intracranial hemorrhage is a major challenge in medicine. Determining the site and type of intracranial hemorrhage is key to treating the patient. The current method of diagnosing intracranial hemorrhage is a CT scan of the brain. Because tissues such as bra...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem that the subtype classification of intracranial hemorrhage is time-consuming and highly dependent on the professionalism of doctors, the present invention provides a subtype classification algorithm of intracranial hemorrhage based on bilinear pooling applied to CT images

The present invention proposes a new CNN-RNN network architecture for subtype classification of ICH. In the CNN module, starting from the characteristics of intracranial CT images, this problem is viewed from a fine-grained perspective, which solves the problems of ResNet and DenseNet networks. Defects on intracranial CT images

Images