CN-116309299-B - Lung image processing method, device, equipment and readable storage medium

CN116309299BCN 116309299 BCN116309299 BCN 116309299BCN-116309299-B

Abstract

The invention discloses a lung image processing method, a device, equipment and a readable storage medium, wherein the method comprises the steps of obtaining a lung image, inputting the lung image into a texture enhancement network to divide the lung image so as to obtain a segmentation image characteristic of enhanced texture, and the texture enhancement network comprises a fine grain texture enhancement unit. According to the invention, fine grain texture enhancement can be performed on the lung image according to the texture enhancement network, so that segmented image features of enhanced textures are obtained, the subsequent recognition of the segmented image features is facilitated, and the recognition accuracy of lung lesions is improved.

Inventors

HUANG HUA
HU JIESI
YANG YANWU
YE CHENFEI
LV HAIYAN

Assignees

深圳市第三人民医院

Dates

Publication Date: 20260512
Application Date: 20230103

Claims (7)

1. A lung image processing method, characterized in that the lung image processing method comprises the following steps: Acquiring a lung image; inputting the lung image into a texture enhancement network to segment the lung image and obtain segmented image features of enhanced textures, wherein the texture enhancement network comprises a fine grain texture enhancement unit; The texture enhancement network further comprises an encoder and a decoder, wherein the lung image is input into the texture enhancement network to segment the lung image, and the step of obtaining segmented image features of the enhanced texture comprises the following steps: inputting the lung image into a fine grain texture enhancement unit to obtain a first image feature; determining the segmented image features from the first image feature, an encoder, the fine grain texture enhancement unit, and a decoder; The fine grain texture enhancement unit comprises a first convolution layer, PReLU layers, a channel and an attention mechanism module which are sequentially connected, the fine grain texture enhancement unit further comprises a second convolution layer, and the step of inputting the lung image into the fine grain texture enhancement unit to obtain a first image feature comprises the following steps: Inputting the lung image features into a first convolution layer, a PReLU layer, a channel and an attention mechanism module which are connected in sequence to obtain second image features, wherein the first convolution layer is initialized by using a high-pass filter and adopts 7x7 convolution, and the calculation formula of the high-pass filter is as follows: wherein the output of F represents the value of a particular pixel, x and y represent the position of the pixel, K represents the size of the filter, and K is set to 7; performing stitching operation on the second image feature and the lung image feature to obtain a third image feature; Inputting the third image feature into a second convolution layer to obtain a first image feature; The encoder comprises a first residual unit, a second residual unit, a third residual unit, a fourth residual unit and a fifth residual unit, wherein the step of determining the split image features according to the first image features, the encoder, the fine grain texture enhancement unit and the decoder comprises the steps of: Inputting the first image feature into the first residual unit to obtain a fourth image feature; Inputting the fourth image feature into the fine grain texture enhancement unit to obtain a fifth image feature; Inputting the fifth image feature into the second residual unit to obtain a sixth image feature; inputting the sixth image feature into the third residual unit to obtain a seventh image feature; inputting the seventh image feature into the fourth residual unit to obtain an eighth image feature; Inputting the eighth image feature into the fifth residual unit to obtain a ninth image feature; Determining the segmented image features from the fifth image feature, sixth image feature, seventh image feature, eighth image feature, ninth image feature, and decoder; Wherein the number of channels of the first residual unit is 248, the number of channels of the second residual unit is 248, the number of channels of the third residual unit is 112, the number of channels of the fourth residual unit is 112, and the number of channels of the fifth residual unit is 112.
2. The pulmonary image processing method according to claim 1, wherein the first residual unit includes a third convolution layer and a fourth convolution layer, and the step of inputting the first image feature into the first residual unit to obtain a fourth image feature includes: Inputting the first image characteristics into a third convolution layer to obtain a first output result, wherein the third convolution layer comprises a convolution layer with a step length of a first preset value, an instance normalization layer and a PReLU layer; inputting the first output result into a fourth convolution layer to obtain a second output result, wherein the fourth convolution layer comprises a convolution with a step length of a second preset value, an instance normalization layer and a PReLU layer; and inputting the first image characteristic into convolution with the step length being a third preset value to obtain a third output result, and performing matrix addition operation on the second output result and the third output result to obtain a fourth image characteristic.
3. The pulmonary image processing method according to claim 1, wherein the decoder includes a first upsampling unit, a second upsampling unit, a third upsampling unit, and a fourth upsampling unit, the determining the segmented image feature based on the fifth image feature, the sixth image feature, the seventh image feature, the eighth image feature, the ninth image feature, and the decoder includes: performing stitching operation on the ninth image feature and the eighth image feature to obtain a tenth image feature, and inputting the tenth image feature to a first upsampling unit to obtain an eleventh image feature; Performing a stitching operation on the eleventh image feature and the seventh image feature to obtain a twelfth image feature, and inputting the twelfth image feature to a second upsampling unit to obtain a thirteenth image feature; Performing stitching operation on the thirteenth image feature and the sixth image feature to obtain a fourteenth image feature, and inputting the fourteenth image feature into a third upsampling unit to obtain a fifteenth image feature; And performing a stitching operation on the fifteenth image feature and the fifth image feature to obtain a sixteenth image feature, and inputting the sixteenth image feature into a fourth upsampling unit to obtain the segmented image feature.
4. The pulmonary image processing method according to claim 3, wherein the first up-sampling unit includes a fifth convolution layer and a sixth convolution layer, and the step of inputting the tenth image feature to the first up-sampling unit to obtain an eleventh image feature includes: Inputting the tenth image feature to a fifth convolution layer to obtain a fourth output result, wherein the fifth convolution layer comprises a deconvolution layer, an instance normalization layer and a PReLU layer with a step length of a fourth preset value; inputting the fourth output result into a sixth convolution layer to obtain a fifth output result, wherein the sixth convolution layer comprises a convolution with a step length of a fifth preset value, an example normalization layer and a PReLU layer; And determining a sixth output result of the deconvolution of the tenth image feature input to a fifth convolution layer, and performing matrix addition operation on the sixth output result and the fifth output result to obtain an eleventh image feature.
5. A lung image processing apparatus, the lung image processing apparatus comprising: the acquisition module is used for acquiring lung images; The segmentation module is used for inputting the lung image into a texture enhancement network to segment the lung image and obtain segmented image features of enhanced textures, wherein the texture enhancement network comprises a fine grain texture enhancement unit; the texture enhancement network further comprises an encoder and a decoder, wherein the segmentation module is further used for inputting the lung image into a fine grain texture enhancement unit so as to obtain a first image characteristic; determining the segmented image features from the first image feature, an encoder, the fine grain texture enhancement unit, and a decoder; The fine grain texture enhancement unit comprises a first convolution layer, PReLU layers, a channel and an attention mechanism module which are sequentially connected, the fine grain texture enhancement unit further comprises a second convolution layer, the segmentation module is further used for inputting lung image features into the first convolution layer, the PReLU layers, the channel and the attention mechanism module which are sequentially connected to obtain second image features, the first convolution layer is initialized by using a high-pass filter and adopts 7x7 convolution, and the calculation formula of the high-pass filter is as follows: wherein the output of F represents the value of a particular pixel, x and y represent the position of the pixel, K represents the size of the filter, and K is set to 7; performing stitching operation on the second image feature and the lung image feature to obtain a third image feature; Inputting the third image feature into a second convolution layer to obtain a first image feature; The encoder comprises a first residual error unit, a second residual error unit, a third residual error unit, a fourth residual error unit and a fifth residual error unit, wherein the segmentation module is further used for inputting the first image characteristic into the first residual error unit so as to obtain a fourth image characteristic; Inputting the fourth image feature into the fine grain texture enhancement unit to obtain a fifth image feature; Inputting the fifth image feature into the second residual unit to obtain a sixth image feature; inputting the sixth image feature into the third residual unit to obtain a seventh image feature; inputting the seventh image feature into the fourth residual unit to obtain an eighth image feature; Inputting the eighth image feature into the fifth residual unit to obtain a ninth image feature; Determining the segmented image features from the fifth image feature, sixth image feature, seventh image feature, eighth image feature, ninth image feature, and decoder; Wherein the number of channels of the first residual unit is 248, the number of channels of the second residual unit is 248, the number of channels of the third residual unit is 112, the number of channels of the fourth residual unit is 112, and the number of channels of the fifth residual unit is 112.
6. A lung image processing device, characterized in that it comprises a memory, a processor and a lung image processing program stored on the memory and executable on the processor, which lung image processing program, when executed by the processor, realizes the steps of the lung image processing method according to any of claims 1-4.
7. A computer-readable storage medium, on which a lung image processing program is stored, which when executed by a processor implements the steps of the lung image processing method according to any of claims 1 to 4.

Description

Lung image processing method, device, equipment and readable storage medium Technical Field The present invention relates to the field of lung image processing technologies, and in particular, to a lung image processing method, device, apparatus and readable storage medium. Background CT imaging of the human lung is considered an important tool for diagnosing and monitoring lung infections. Studies have shown that lesion size and severity can be assessed from chest CT images to assess disease progression and subsequent treatment. However, manually identifying these affected areas is very inefficient, typically requiring several hours to complete identification of individual patients, and thus it is highly necessary to build reliable artificial intelligence assisted labeling tools to increase the efficiency of identification. However, for some lung lesions with high heterogeneity and unclear boundaries, the recognition effect based on artificial intelligence is also poor. The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present invention and is not intended to represent an admission that the foregoing is prior art. Disclosure of Invention The invention mainly aims to provide a lung image processing method, a device, equipment and a readable storage medium, aiming at poor lung focus identification effect based on artificial intelligence. In order to achieve the above object, the present invention provides a lung image processing method, which includes the steps of: Acquiring a lung image; inputting the lung image into a texture enhancement network to segment the lung image and obtain segmented image features of enhanced textures, wherein the texture enhancement network comprises a fine grain texture enhancement unit. Further, the texture enhancement network further comprises an encoder and a decoder, the step of inputting the lung image into the texture enhancement network to segment the lung image and obtain segmented image features of the enhanced texture comprises: inputting the lung image into a fine grain texture enhancement unit to obtain a first image feature; the segmented image feature is determined from the first image feature, an encoder, the fine grain texture enhancement unit, and a decoder. Further, the fine grain texture enhancement unit comprises a first convolution layer, PReLU layers, a channel and an attention mechanism module which are connected in sequence, the fine grain texture enhancement unit further comprises a second convolution layer, and the step of inputting the lung image into the fine grain texture enhancement unit to obtain a first image feature comprises the following steps: inputting the lung image features to a first convolution layer, PReLU layers, a channel and an attention mechanism module which are connected in sequence to obtain second image features, wherein the first convolution layer is initialized by using a high-pass filter; performing stitching operation on the second image feature and the lung image feature to obtain a third image feature; The third image feature is input into a second convolution layer to obtain a first image feature. Further, the encoder comprises a first residual unit, a second residual unit, a third residual unit, a fourth residual unit and a fifth residual unit, and the step of determining the segmented image feature from the first image feature, the encoder, the fine grain texture enhancement unit and the decoder comprises: Inputting the first image feature into the first residual unit to obtain a fourth image feature; Inputting the fourth image feature into the fine grain texture enhancement unit to obtain a fifth image feature; Inputting the fifth image feature into the second residual unit to obtain a sixth image feature; inputting the sixth image feature into the third residual unit to obtain a seventh image feature; inputting the seventh image feature into the fourth residual unit to obtain an eighth image feature; Inputting the eighth image feature into the fifth residual unit to obtain a ninth image feature; the segmented image features are determined from the fifth image feature, the sixth image feature, the seventh image feature, the eighth image feature, the ninth image feature, and the decoder. Further, the first residual unit comprises a third convolution layer and a fourth convolution layer, and the step of inputting the first image feature into the first residual unit to obtain a fourth image feature comprises the steps of: Inputting the first image characteristics into a third convolution layer to obtain a first output result, wherein the third convolution layer comprises a convolution layer with a step length of a first preset value, an instance normalization layer and a PReLU layer; inputting the first output result into a fourth convolution layer to obtain a second output result, wherein the fourth convolution layer comprises a convolution with a step length