CN115619889A - Multi-feature fusion photoacoustic image reconstruction method suitable for annular array - Google Patents
Multi-feature fusion photoacoustic image reconstruction method suitable for annular array Download PDFInfo
- Publication number
- CN115619889A CN115619889A CN202211402980.9A CN202211402980A CN115619889A CN 115619889 A CN115619889 A CN 115619889A CN 202211402980 A CN202211402980 A CN 202211402980A CN 115619889 A CN115619889 A CN 115619889A
- Authority
- CN
- China
- Prior art keywords
- layer
- multiplied
- conv1
- feature
- extraction module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 28
- 230000004927 fusion Effects 0.000 title claims abstract description 21
- 238000000605 extraction Methods 0.000 claims abstract description 46
- 238000010586 diagram Methods 0.000 claims abstract description 17
- 238000005070 sampling Methods 0.000 claims abstract description 10
- 238000012952 Resampling Methods 0.000 claims abstract description 5
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 238000010606 normalization Methods 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 15
- 230000004913 activation Effects 0.000 claims description 5
- 238000013461 design Methods 0.000 abstract description 2
- 238000002474 experimental method Methods 0.000 abstract description 2
- 238000003384 imaging method Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 2
- 238000003325 tomography Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010895 photoacoustic effect Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/006—Inverse problem, transformation from projection-space into object-space, e.g. transform methods, back-projection, algebraic methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2211/00—Image generation
- G06T2211/40—Computed tomography
- G06T2211/421—Filtered back projection [FBP]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Mathematical Analysis (AREA)
- Algebra (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Physics (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a multi-feature fusion photoacoustic image reconstruction method suitable for an annular array, which comprises the following steps of: step one, preprocessing data; step two, designing a signal domain feature extraction module; step three, designing an image domain feature extraction module; and fourthly, overlapping and outputting the characteristic diagram. The photoacoustic reconstruction method disclosed by the invention integrates the original photoacoustic signal, the first derivative information of the photoacoustic signal to time and the low-quality image reconstructed by the traditional reconstruction algorithm, and experiments prove that the method has excellent performance and can realize the rapid and high-quality photoacoustic image reconstruction under the limited viewing angle. The method has flexible design, carries out interpolation resampling in the characteristic extraction process, automatically adjusts the signal dimension, and is suitable for the photoacoustic signals with different sampling frequencies.
Description
Technical Field
The invention belongs to the technical field of artificial intelligence, and relates to a photoacoustic image reconstruction method, in particular to a multi-input characteristic fusion photoacoustic image reconstruction method for annular array photoacoustic tomography.
Background
Photoacoustic tomography (PAT) is an emerging noninvasive biomedical imaging method based on photoacoustic effect, has the characteristics of large imaging depth and high contrast, and can be used for medical diagnosis and treatment assistance. However, due to the limitation of the equipment cost and the requirement of reconstruction time, the existing ring array transducer photoacoustic imaging system has difficulty in simultaneously considering the image quality and the imaging speed. Conventional PAT reconstruction algorithms, such as filtered backprojection and temporal inversion algorithms, have been widely used for photoacoustic image reconstruction. But these reconstruction algorithms will generate distorted images containing many artifacts under limited viewing angles. Therefore, a fast and high-quality photoacoustic image reconstruction method is found to improve the quality of the reconstructed image of undersampled data under a limited visual angle, and the method has great significance for promoting clinical transformation and application of the PAT technology.
Disclosure of Invention
The invention aims to provide a multi-feature fusion photoacoustic image reconstruction method suitable for an annular array, which takes an original photoacoustic signal and a traditional reconstructed image as input, and simultaneously takes a photoacoustic physical model (namely a first-order partial derivative of the signal to time) as prior information to guide a reconstruction process so as to realize rapid and high-quality photoacoustic image reconstruction under a limited viewing angle.
The purpose of the invention is realized by the following technical scheme:
a multi-feature fusion photoacoustic image reconstruction method suitable for an annular array comprises the following steps:
step one, data preprocessing
The method comprises the steps of setting the radius of a circular array distribution circle as r and the sampling rate as f s The propagation speed of the sound wave is 1540m/s, and the original photoacoustic signal p (d) is obtained by sampling with an ultrasonic transducer with n array elements i ,t)，d i The distance from an ultrasonic transducer array element i (i belongs to n) to a pixel point, and t is time;
step two, obtaining an original photoacoustic signal p (d) based on the following formula i T) first derivative with respect to time t
And storing:
wherein Δ t =1/f s ；
Thirdly, generating a low-quality reconstruction image of 128 multiplied by 128 by an original photoacoustic signal through a traditional reconstruction algorithm and storing the low-quality reconstruction image;
step four, normalization processing is carried out on the original photoacoustic signal and the photoacoustic signal first-order derivative based on the following formula:
where p is a set of signal samples, p (i) represents the original photoacoustic signal of array element i,
representing the photoacoustic signal corresponding to the array element i after normalization processing;
step two, designing a signal domain feature extraction module
The signal domain feature extraction module comprises an original photoacoustic signal feature extraction module, a photoacoustic signal first derivative feature extraction module, a feature fusion module and a fusion feature processing module, wherein:
the original photoacoustic signal feature extraction module comprises a first layer Conv1 × 7, a second layer Resblock and a third layer Conv1 × 7, the original photoacoustic signal is input into the first layer Conv1 × 7 after normalization processing, the input data dimension of the first layer Conv1 × 7 is b × 1 × n × s, and a feature diagram with the dimension of b × c × n × s is output to the second layer Resblock; the output dimension of the second layer of the Resblock is b multiplied by c multiplied by n multiplied by s to a third layer of Conv1 multiplied by 7; the third layer Conv1 × 7 outputs a feature map A with b × c × n × (s/2) in dimension;
the photoacoustic signal first-order derivative feature extraction module comprises a first layer Conv1 × 7, a second layer Resblock and a third layer Conv1 × 7, wherein the photoacoustic signal first-order derivative is input into the first layer Conv1 × 7 after normalization processing, the input dimension of the first layer Conv1 × 7 is b × 1 × n × s, and a feature map with the dimension of b × c × n × s is output to the second layer Resblock; the output dimension of the second layer of the Resblock is b multiplied by c multiplied by n multiplied by s to a third layer of Conv1 multiplied by 7; the third layer Conv1 × 7 outputs a feature map B with B × c × n × (s/2) dimensions;
the feature fusion module fuses a B × C × n × (s/2) feature map A output by an original photoacoustic signal feature extraction module and a B × C × n × (s/2) feature map B output by a photoacoustic signal first-derivative feature extraction module into a B × 2C × n × (s/2) feature map according to channels by using a Concat mode, and outputs a B × 2C × 64 × 512 feature map C after bilinear interpolation and resampling and size adjustment;
the fusion feature extraction module comprises a fourth layer of Resblock, a fifth layer of Conv1 x 3, a sixth layer of Conv1 x 3, a seventh layer of Resblock + SEblock, an eighth layer of Conv1 x 3 and a ninth layer of FC all-connected layer, wherein the fourth layer of Resblock has an input dimension of b x 2C x 64 x 512 of a feature diagram C, and outputs a feature diagram with an output dimension of b x 4C x 64 x 256 to the fifth layer of Conv1 x 3; a fifth layer Conv1 × 3 to a sixth layer Conv1 × 3 of feature maps having output dimensions b × 2c × 32 × 256; the sixth layer Conv1 x 3 outputs characteristic graphs with dimensions of b x c x 32 x 256 to the seventh layer Resblock + SEblock; a feature diagram with the output dimensions of b × c × 32 × 256 of the seventh layer Resblock + SEblock to the eighth layer Conv1 × 3; the eighth layer Conv1 × 3 outputs a feature map with output dimensions of b × (c/2) × 32 × 128 to the ninth layer FC full connection layer; a ninth FC full-connection layer outputs a feature map D with dimensions b multiplied by 1 multiplied by 128 through dimension reconstruction;
step three, designing an image domain feature extraction module
The image domain feature extraction module comprises a 6-layer Conv3 x 3 convolution structure, the image domain feature extraction module takes a b x 1 x 128 low-quality reconstructed image as input, and outputs a feature map E with b x 1 x 128 of output dimension;
step four, feature map superposition output
And performing summation operation sigma on the feature map D output by the signal domain feature extraction module and the feature map E output by the image domain feature extraction module to obtain a b multiplied by 1 multiplied by 128 photoacoustic reconstruction image.
Compared with the prior art, the invention has the following advantages:
1. the photoacoustic reconstruction method disclosed by the invention integrates the original photoacoustic signal, the first derivative information of the photoacoustic signal to time and the low-quality image reconstructed by the traditional reconstruction algorithm, and experiments prove that the method has excellent performance and can realize the rapid and high-quality photoacoustic image reconstruction under the limited viewing angle.
2. The method has flexible design, carries out interpolation resampling in the characteristic extraction process, automatically adjusts the signal dimension, and is suitable for the photoacoustic signals with different sampling frequencies.
Drawings
FIG. 1 is a structure of a reconstructed model according to the present invention;
FIG. 2 is a diagram illustrating the specific operation of Resblock;
FIG. 3 is a diagram illustrating the detailed operation of SEblock;
fig. 4 is an example of reconstructing an image based on geometry.
Detailed Description
The technical solutions of the present invention are further described below with reference to the drawings, but the present invention is not limited thereto, and any modifications or equivalent substitutions may be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
The invention provides a multi-feature fusion photoacoustic image reconstruction method suitable for an annular array, which comprises the following steps of:
step one, data preprocessing
Let the radius of the circular array distribution circle be r and the sampling rate be f s The acoustic wave propagation speed is 1540m/s, and the original photoacoustic signal p (d) is obtained by sampling with an n-array element ultrasonic transducer i ,t)，d i The distance from the ultrasonic transducer element i (i ∈ n) to the pixel point is obtained.
1. First partial derivative of photoacoustic signal with respect to time
Original photoacoustic signal p (d) i T) first partial derivative with respect to time t
Can be expressed as:
wherein, Δ t =1/f s . And (3) obtaining a first derivative of the original photoacoustic signal to time based on the formula (1) and storing the first derivative.
2. Low quality reconstructed images
The original photoacoustic signal is subjected to a traditional reconstruction algorithm (such as filtered back projection and time reversal) to generate a 128 x 128 low-quality reconstructed image and is stored.
3. Normalization process
In order to accelerate the convergence of the reconstruction algorithm, before the original photoacoustic signal and the prior information of the first derivative of the photoacoustic signal are input into the network, the normalization processing is performed as shown in formula (2):
step two, designing a signal domain feature extraction module
As shown in fig. 1, the signal domain feature extraction module mainly includes an original photoacoustic signal feature extraction module (module 1), a photoacoustic signal first-order derivative feature extraction module (module 2), a feature fusion module (module 3), and a fusion feature processing module (module 4). Wherein the Conv1 × 7 layer generally includes a convolution kernel size of 1 × 7 convolution → Instance Normalization (IN) → ReLU activation function; the Conv1 × 3 layer generally includes a convolution kernel size of 1 × 3 convolution → Instance Normalization (IN) → ReLU activation function; the specific operation of Resblock is shown in fig. 2, and the specific operation of SEblock is shown in fig. 3. Meanwhile, the convolution channel number c can be adaptively adjusted according to the sample size and the operating environment, and is preferably selected to be 16, 32 or 64.
Original photoacoustic signal feature extraction module (module 1): including a first layer Conv1 × 7, a second layer Resblock, and a third layer Conv1 × 7. The original photoacoustic signals are input into a first layer Conv1 × 7 through normalization processing, the dimension of input data of the first layer Conv1 × 7 is b × 1 × n × s, namely the size of data loaded in each batch is b, n corresponds to the number of array elements, s represents the length of the acquired signals, the size of c convolution kernels used is 1 × 7, the step size is (1,1), padding is (0,3), and a feature map with the dimension of b × c × n × s is output to a second layer Resblock; the output dimension of the second layer of the Resblock is b multiplied by c multiplied by n multiplied by s to a third layer of Conv1 multiplied by 7; the convolution kernel used by the third layer Conv1 × 7 has a size of 1 × 7, a step size of (1,2), a padding of (0,3), and an output dimension of b × c × n × (s/2).
Photoacoustic signal first derivative feature extraction module (module 2): including a first layer Conv1 × 7, a second layer Resblock, and a third layer Conv1 × 7. The photoacoustic first-order derivative is input into a first layer Conv1 × 7 through normalization processing, the input dimension of the first layer Conv1 × 7 is consistent with that of an original photoacoustic signal, namely b × 1 × n × s, the size of c used convolution kernels is 1 × 7, the step size is (1,1), the padding is (0,3), and a feature map with the dimension of b × c × n × s is output to a second layer Resblock; outputting a characteristic diagram with the dimension of b multiplied by c multiplied by n multiplied by s to a third layer Conv1 multiplied by 7 from the second layer Resblock; the convolution kernel used by the third layer Conv1 × 7 has a size of 1 × 7, a step size of (1,2), a padding of (0,3), and an output dimension of B × c × n × (s/2).
Feature fusion module (module 3): and performing feature fusion by using a Concat method, and adjusting the dimensions of the fused feature map to be b × 2c × 64 × 512. B multiplied by C multiplied by n (s/2) characteristic diagram A output by an original photoacoustic signal characteristic extraction module and B multiplied by C multiplied by n (s/2) characteristic diagram B output by a photoacoustic signal first derivative characteristic extraction module are fused into a B multiplied by 2C multiplied by n (s/2) characteristic diagram according to channels, and B multiplied by 2C multiplied by 64 multiplied by 512 characteristic diagram C is output through bilinear interpolation resampling and size adjustment.
Fusion feature extraction module (module 4): comprising a fourth layer Resblock, a fifth layer Conv1 × 3, a sixth layer Conv1 × 3, a seventh layer Resblock + SEblock, an eighth layer Conv1 × 3 and a ninth layer FC fully connected layer. A feature map C with b × 2C × 64 × 512 is input into the fourth layer Resblock, and a feature map with b × 4C × 64 × 256 is output to the fifth layer Conv1 × 3; the convolution kernel size used by the fifth layer Conv1 × 3 is 1 × 3, the step size is (2,1), the padding is (0,1), and the output dimension is b × 2c × 32 × 256 to the sixth layer Conv1 × 3; the convolution kernel size used by the sixth layer Conv1 × 3 is 1 × 3, the step size is (1,1), the padding is (0,1), and the output dimensions are b × c × 32 × 256 to the seventh layer Resblock + SEblock; a feature diagram with the output dimensions of b × c × 32 × 256 of the seventh layer Resblock + SEblock to the eighth layer Conv1 × 3; the convolution kernel size used by the eighth layer Conv1 × 3 is 1 × 3, the step size is (1,2), the padding is (0,1), and the output dimension is b × (c/2) × 32 × 128 to the ninth layer FC fully-connected layer; and the ninth FC full-connection layer outputs a feature map D with dimensions b multiplied by 1 multiplied by 128 after dimension reconstruction.
Step three, designing an image domain feature extraction module
No pooling operation is employed on the input b × 1 × 128 × 128 low quality reconstructed image for rich details and global texture features. As shown in fig. 1, the image domain feature extraction module (module 5) mainly comprises a 6-layer Conv3 × 3 convolution structure. The Conv3 × 3 layer generally includes a convolution kernel size of 3 × 3 convolution → Batch Normalization (BN) → ReLU activation function. And outputting a feature map E with dimensions b multiplied by 1 multiplied by 128 after the feature extraction of the image domain.
Step four, feature map superposition output
And performing summation operation sigma on the feature map D output by the signal domain feature extraction module and the feature map E output by the image domain feature extraction module to obtain a b multiplied by 1 multiplied by 128 photoacoustic reconstruction image.
Example (b):
the method provided by the invention is used for reconstructing the geometric body, the radius of a circular array distribution circle is 55mm, the sampling rate is 10.267MHz, the sound wave propagation speed is 1540m/s, and a 64-array-element ultrasonic transducer under a 180-degree visual angle is used for sampling. Meanwhile, the original photoacoustic signal with the dimension of 1 × 64 × 1024, the first derivative of the photoacoustic signal with the dimension of 1 × 64 × 1024 with respect to time and the low-quality photoacoustic image with the dimension of 1 × 128 × 128 are obtained through data preprocessing.
The data set is divided into a training set and a testing set, and the data volume size b of each batch load is 4,c and 32. The training set comprises 1400 pieces of single-type geometry (circle, circular arc, rectangle and line) and 2100 pieces of multi-type geometry (random combination of any single-type geometry); the test set comprises 200 single-class geometric bodies (circles, circular arcs, rectangles and lines) and 300 multi-class geometric bodies (random combination of any single-class geometric bodies). The reconstruction model provided by the invention can complete the reconstruction of a single high-quality photoacoustic image in about 1s, and as shown in table 1, the reconstruction performance PSNR in a test set is more than 28dB, and SSIM is more than 0.65.
TABLE 1 evaluation index based on geometry simulation dataset
The comparison between the reconstructed image of the network model proposed by the present invention and the reconstructed image of the conventional reconstruction method TR is shown in fig. 4.
Claims (5)
1. A multi-feature fusion photoacoustic image reconstruction method suitable for a ring array is characterized by comprising the following steps:
step one, data preprocessing
The method comprises the steps of setting the radius of a circular array distribution circle as r and the sampling rate as f s The acoustic wave propagation speed is 1540m/s, and the original photoacoustic signal p (d) is obtained by sampling with an n-array element ultrasonic transducer i ,t)，d i The distance from an ultrasonic transducer array element i to a pixel point is represented by i belonging to n;
step two, obtaining an original photoacoustic signal p (d) based on the following formula i T) first derivative with respect to time
And storing:
wherein, Δ t =1/f s ；
Thirdly, generating a low-quality reconstruction image of 128 multiplied by 128 by an original photoacoustic signal through a traditional reconstruction algorithm and storing the low-quality reconstruction image;
step four, performing normalization processing on the original photoacoustic signal and the first-order derivative of the photoacoustic signal;
step two, designing a signal domain feature extraction module
The signal domain feature extraction module comprises an original photoacoustic signal feature extraction module, a photoacoustic signal first derivative feature extraction module, a feature fusion module and a fusion feature processing module, wherein:
the original photoacoustic signal feature extraction module comprises a first layer Conv1 × 7, a second layer Resblock and a third layer Conv1 × 7, the original photoacoustic signal is input into the first layer Conv1 × 7 after normalization processing, the input data dimension of the first layer Conv1 × 7 is b × 1 × n × s, and a feature diagram with the dimension of b × c × n × s is output to the second layer Resblock; the output dimension of the second layer of the Resblock is b multiplied by c multiplied by n multiplied by s to a third layer of Conv1 multiplied by 7; the third layer Conv1 × 7 outputs a feature map A with b × c × n × (s/2) in dimension;
the photoacoustic signal first-order derivative feature extraction module comprises a first layer Conv1 × 7, a second layer Resblock and a third layer Conv1 × 7, wherein the photoacoustic signal first-order derivative is input into the first layer Conv1 × 7 after normalization processing, the input dimension of the first layer Conv1 × 7 is b × 1 × n × s, and a feature map with the dimension of b × c × n × s is output to the second layer Resblock; the output dimension of the second layer of the Resblock is b multiplied by c multiplied by n multiplied by s to a third layer of Conv1 multiplied by 7; the third layer Conv1 × 7 outputs a feature map B with B × c × n × (s/2) dimensions;
the feature fusion module fuses a B multiplied by C multiplied by n multiplied by (s/2) feature map A output by an original photoacoustic signal feature extraction module and a B multiplied by C multiplied by n multiplied by (s/2) feature map B output by a photoacoustic signal first derivative feature extraction module into a B multiplied by 2C multiplied by n multiplied by (s/2) feature map according to channels by using a Concat mode, and outputs a B multiplied by 2C multiplied by 64 multiplied by 512 feature map C after bilinear interpolation and resampling and size adjustment;
the fusion feature extraction module comprises a fourth layer of Resblock, a fifth layer of Conv1 x 3, a sixth layer of Conv1 x 3, a seventh layer of Resblock + SEblock, an eighth layer of Conv1 x 3 and a ninth layer of FC all-connected layer, wherein the fourth layer of Resblock has an input dimension of b x 2C x 64 x 512 of a feature diagram C, and outputs a feature diagram with an output dimension of b x 4C x 64 x 256 to the fifth layer of Conv1 x 3; a fifth layer Conv1 × 3 to a sixth layer Conv1 × 3 having characteristic diagrams with b × 2c × 32 × 256 output dimensions; the sixth layer Conv1 x 3 outputs characteristic graphs with dimensions of b x c x 32 x 256 to the seventh layer Resblock + SEblock; the seventh layer Resblock + SEblock outputs a feature map with dimensions b × c × 32 × 256 to the eighth layer Conv1 × 3; the eighth layer Conv1 × 3 outputs a feature map having an output dimension of b × (c/2) × 32 × 128 to the ninth layer FC full connection layer; a ninth FC full connection layer outputs a feature map D with the dimension of b multiplied by 1 multiplied by 128 through dimension reconstruction;
step three, designing an image domain feature extraction module
The image domain feature extraction module comprises a 6-layer Conv3 x 3 convolution structure, the image domain feature extraction module takes a b x 1 x 128 low-quality reconstructed image as input, and outputs a feature map E with b x 1 x 128 of output dimension;
step four, feature map superposition output
And performing summation operation sigma on the feature map D output by the signal domain feature extraction module and the feature map E output by the image domain feature extraction module to obtain the b × 1 × 128 × 128 photoacoustic reconstruction image.
2. The method for reconstructing the multi-feature fused photoacoustic image suitable for use in an annular array according to claim 1, wherein the normalization processing formula is as follows:
wherein,p is a set of signal samples, p (i) represents the original photoacoustic signal of array element i,
and representing the photoacoustic signal corresponding to the array element i after normalization processing.
3. The method of claim 1, wherein the Conv1 x 7 layer comprises a convolution kernel size of 1 x 7 convolution → instance normalization → ReLU activation function.
4. The method of claim 1, wherein the Conv1 x 3 layer comprises convolution kernel size 1 x 3 convolution → instance normalization → ReLU activation function.
5. The method of claim 1, wherein c is selected to be 16, 32 or 64.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211402980.9A CN115619889B (en) | 2022-11-09 | 2022-11-09 | Multi-feature fusion photoacoustic image reconstruction method suitable for annular array |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211402980.9A CN115619889B (en) | 2022-11-09 | 2022-11-09 | Multi-feature fusion photoacoustic image reconstruction method suitable for annular array |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115619889A true CN115619889A (en) | 2023-01-17 |
| CN115619889B CN115619889B (en) | 2023-05-30 |
Family
ID=84877798
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211402980.9A Active CN115619889B (en) | 2022-11-09 | 2022-11-09 | Multi-feature fusion photoacoustic image reconstruction method suitable for annular array |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115619889B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116740219A (en) * | 2023-08-14 | 2023-09-12 | 之江实验室 | Three-dimensional photoacoustic tomography method, device, equipment and readable storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103345770A (en) * | 2013-07-18 | 2013-10-09 | 中国科学院自动化研究所 | Limited viewing angle photoacoustic imaging reestablishing method based on iteration self-adaption weighting |
| CN108577810A (en) * | 2018-03-21 | 2018-09-28 | 华北电力大学（保定） | Solve the intravascular photoacoustic image rebuilding method and system of velocity of sound problem of non-uniform |
| CN112465924A (en) * | 2020-12-11 | 2021-03-09 | 上海科技大学 | Rapid medical image reconstruction method based on multi-feature fusion |
| US20220028128A1 (en) * | 2018-10-25 | 2022-01-27 | Nanjing University | Photoacoustic image reconstruction method for suppressing artifacts |
| CN114332283A (en) * | 2021-12-31 | 2022-04-12 | 之江实验室 | Training method based on double-domain neural network and photoacoustic image reconstruction method |
| CN114692509A (en) * | 2022-04-21 | 2022-07-01 | 南京邮电大学 | Strong noise single photon three-dimensional reconstruction method based on multi-stage degeneration neural network |
| CN115177217A (en) * | 2022-09-09 | 2022-10-14 | 之江实验室 | Photoacoustic signal simulation method and device based on spherical particle light pulse excitation effect |
-
2022
- 2022-11-09 CN CN202211402980.9A patent/CN115619889B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103345770A (en) * | 2013-07-18 | 2013-10-09 | 中国科学院自动化研究所 | Limited viewing angle photoacoustic imaging reestablishing method based on iteration self-adaption weighting |
| CN108577810A (en) * | 2018-03-21 | 2018-09-28 | 华北电力大学（保定） | Solve the intravascular photoacoustic image rebuilding method and system of velocity of sound problem of non-uniform |
| US20220028128A1 (en) * | 2018-10-25 | 2022-01-27 | Nanjing University | Photoacoustic image reconstruction method for suppressing artifacts |
| CN112465924A (en) * | 2020-12-11 | 2021-03-09 | 上海科技大学 | Rapid medical image reconstruction method based on multi-feature fusion |
| CN114332283A (en) * | 2021-12-31 | 2022-04-12 | 之江实验室 | Training method based on double-domain neural network and photoacoustic image reconstruction method |
| CN114692509A (en) * | 2022-04-21 | 2022-07-01 | 南京邮电大学 | Strong noise single photon three-dimensional reconstruction method based on multi-stage degeneration neural network |
| CN115177217A (en) * | 2022-09-09 | 2022-10-14 | 之江实验室 | Photoacoustic signal simulation method and device based on spherical particle light pulse excitation effect |
Non-Patent Citations (2)
| Title |
|---|
| JIADONG ZHANG.ETC: "LIMITED-VIEW PHOTOACOUSTIC IMAGING RECONSTRUCTION WITH DUAL DOMAIN INPUTS BASED ON MUTUAL INFORAMTION", 《IEEE》 * |
| 沈康等: "基于双域神经网络的稀疏视角光声图像重建", 《中国激光》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116740219A (en) * | 2023-08-14 | 2023-09-12 | 之江实验室 | Three-dimensional photoacoustic tomography method, device, equipment and readable storage medium |
| CN116740219B (en) * | 2023-08-14 | 2024-01-09 | 之江实验室 | Three-dimensional photoacoustic tomography method, device, equipment and readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115619889B (en) | 2023-05-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109978778B (en) | Convolutional neural network medical CT image denoising method based on residual learning | |
| Zhou et al. | High spatial–temporal resolution reconstruction of plane-wave ultrasound images with a multichannel multiscale convolutional neural network | |
| CN111445390A (en) | Wide residual attention-based three-dimensional medical image super-resolution reconstruction method | |
| CN111932461B (en) | Self-learning image super-resolution reconstruction method and system based on convolutional neural network | |
| CN110880196B (en) | Tumor photoacoustic image rapid reconstruction method and device based on deep learning | |
| US20210265042A1 (en) | Ultrasound imaging by deep learning and associated devices, systems, and methods | |
| CN115619889A (en) | Multi-feature fusion photoacoustic image reconstruction method suitable for annular array | |
| CN110473150B (en) | CNN medical CT image denoising method based on multi-feature extraction | |
| CN113989125B (en) | Method and device for splicing endoscope images, computer equipment and storage medium | |
| CN112102428B (en) | CT cone beam scanning image reconstruction method, scanning system and storage medium | |
| Zhao et al. | Sparse-view CT reconstruction via generative adversarial networks | |
| Li et al. | Beamforming with deep learning from single plane wave RF data | |
| CN117237473A (en) | Photoacoustic tomography sparse reconstruction method based on diffusion model | |
| CN111178499A (en) | Medical image super-resolution method based on generation countermeasure network improvement | |
| Asgariandehkordi et al. | Deep ultrasound denoising using diffusion probabilistic models | |
| Seetharaman et al. | Retracted: A novel approach in hybrid median filtering for denoising medical images | |
| Huh et al. | Tunable image quality control of 3-D ultrasound using switchable CycleGAN | |
| CN113947662A (en) | Ultrasonic simulation method and system based on medical tomography | |
| CN114332283A (en) | Training method based on double-domain neural network and photoacoustic image reconstruction method | |
| CN117197349A (en) | CT image reconstruction method and device | |
| CN115937345A (en) | Photoacoustic image reconstruction method and device based on deep learning | |
| Dey et al. | Score-based Diffusion Models for Photoacoustic Tomography Image Reconstruction | |
| CN107847141A (en) | For obtaining equipment, the methods and procedures of the information relevant with the position displacement of multiple images data set | |
| Cammarasana et al. | Super-resolution of 2D ultrasound images and videos | |
| Fan et al. | Memory-efficient neural network for non-linear ultrasound computed tomography reconstruction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |