Skull fracture detection and model training method, device, equipment and storage medium

Abstract

The embodiment of the invention discloses a skull fracture detection and model training method, device, equipment and storage medium. The model training method includes: obtaining multiple groups of skull fracture training sample data; based on each group of skull fracture training sample data, training the initial target detection model based on the neural network to generate a skull fracture detection model; wherein, the skull fracture detection model includes a depth The feature extraction main network and the non-candidate frame classification sub-network composed of convolutional neural network and multi-scale target detection network; the feature extraction main network is used to extract features from head medical images, obtain at least one fusion feature layer, and fuse the feature layers The location information of the low-level convolutional feature layer and the semantic information of the high-level convolutional feature layer are obtained; the non-candidate frame classification subnetwork is used to perform convolution operations on at least the last fusion feature layer in each fusion feature layer to obtain the model output result. Through the above technical solution, high-precision detection of skull fractures is realized.

Description

technical field [0001] Embodiments of the present invention relate to medical image processing technology, and in particular to a skull fracture detection and model training method, device, equipment and storage medium. Background technique [0002] Skull fracture is a common disease of craniocerebral trauma, and cranial brain computed tomography (CT) images are now the main basis for the diagnosis of skull fractures. [0003] At present, there are mainly two methods for diagnosing skull fractures using cranial CT images: one is to identify the skull fractures in the CT images by reading the images with human eyes; the other is to process and target the cranial CT images for automatic identification One of the skull fractures. [0004] However, the way of human eye recognition is likely to cause missed diagnosis because it depends on the doctor's experience and is prone to fatigue. In the automatic identification method, due to the different sizes of skull fractures, some ...

AI Technical Summary

Problems solved by technology

[0004] However, the way of human eye recognition, because it depends on the doctor's experience and is prone to fatigue, may easily lead to missed diagnosis.

In the automatic identification method, due to the different sizes of skull fractures, some are relatively large-scale fractures, and some are not obvious comminuted fractures, which limits the applicability of existing target detection algorithms and reduces the detection of skull fractures. precision

Images