CN112164028A - Pituitary adenoma magnetic resonance image positioning diagnosis method and device based on artificial intelligence - Google Patents
Pituitary adenoma magnetic resonance image positioning diagnosis method and device based on artificial intelligence Download PDFInfo
- Publication number
- CN112164028A CN112164028A CN202010907454.2A CN202010907454A CN112164028A CN 112164028 A CN112164028 A CN 112164028A CN 202010907454 A CN202010907454 A CN 202010907454A CN 112164028 A CN112164028 A CN 112164028A
- Authority
- CN
- China
- Prior art keywords
- pituitary
- magnetic resonance
- resonance image
- image
- detected
- 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.)
- Pending
Links
- 208000007913 Pituitary Neoplasms Diseases 0.000 title claims abstract description 161
- 206010061538 Pituitary tumour benign Diseases 0.000 title claims abstract description 133
- 201000005746 Pituitary adenoma Diseases 0.000 title claims abstract description 131
- 208000021310 pituitary gland adenoma Diseases 0.000 title claims abstract description 131
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 29
- 238000003745 diagnosis Methods 0.000 title claims abstract description 29
- 230000001817 pituitary effect Effects 0.000 claims abstract description 297
- 238000012549 training Methods 0.000 claims abstract description 110
- 238000007781 pre-processing Methods 0.000 claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 20
- 238000004364 calculation method Methods 0.000 claims abstract description 16
- 238000013527 convolutional neural network Methods 0.000 claims description 39
- 238000004422 calculation algorithm Methods 0.000 claims description 18
- 230000004807 localization Effects 0.000 claims description 14
- 208000010916 pituitary tumor Diseases 0.000 claims description 11
- 238000013528 artificial neural network Methods 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 36
- 238000010586 diagram Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 11
- 238000012360 testing method Methods 0.000 description 11
- 230000004913 activation Effects 0.000 description 10
- 238000002595 magnetic resonance imaging Methods 0.000 description 9
- 230000006872 improvement Effects 0.000 description 8
- 238000011176 pooling Methods 0.000 description 7
- 238000002372 labelling Methods 0.000 description 6
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 238000007917 intracranial administration Methods 0.000 description 3
- 238000010030 laminating Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- HUPNQNOWXCVQSW-UHFFFAOYSA-N 2h-pyran-4-carboxamide Chemical compound NC(=O)C1=CCOC=C1 HUPNQNOWXCVQSW-UHFFFAOYSA-N 0.000 description 2
- 206010062767 Hypophysitis Diseases 0.000 description 2
- 238000005481 NMR spectroscopy Methods 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 210000005013 brain tissue Anatomy 0.000 description 2
- 210000004907 gland Anatomy 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 210000003635 pituitary gland Anatomy 0.000 description 2
- 206010052399 Neuroendocrine tumour Diseases 0.000 description 1
- 210000003484 anatomy Anatomy 0.000 description 1
- 238000011888 autopsy Methods 0.000 description 1
- 210000001638 cerebellum Anatomy 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 210000000750 endocrine system Anatomy 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 229940088597 hormone Drugs 0.000 description 1
- 239000005556 hormone Substances 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 208000016065 neuroendocrine neoplasm Diseases 0.000 description 1
- 201000011519 neuroendocrine tumor Diseases 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 210000003625 skull Anatomy 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Abstract
The invention discloses an artificial intelligence-based pituitary adenoma magnetic resonance image positioning diagnosis method, which comprises the following steps: acquiring at least one pituitary magnetic resonance image of a person to be detected; preprocessing each pituitary magnetic resonance image of the person to be detected, and stacking to obtain a corresponding magnetic resonance image to be detected; inputting the magnetic resonance image to be detected into a pituitary adenoma recognition model obtained by pre-training so that the pituitary adenoma recognition model can calculate the magnetic resonance image to be detected; and outputting the classification result of whether the magnetic resonance image to be detected contains pituitary adenoma or not. The invention also discloses a corresponding processing device, and by implementing the embodiment of the invention, the calculation and analysis are carried out on the magnetic resonance image to be detected through the pre-established and trained pituitary adenoma recognition model, the classification result of whether the pituitary adenoma exists is output, and the processing efficiency and the processing precision of the pituitary magnetic resonance image are effectively improved.
Description
Technical Field
The invention relates to the technical field of digital image processing, in particular to a pituitary adenoma magnetic resonance image positioning diagnosis method and device based on artificial intelligence.
Background
The pituitary gland, which is the main gland that supervises other glands of the endocrine system and controls hormone levels, is a common neuroendocrine tumor. According to past autopsy and imaging studies, the prevalence of pituitary adenomas is about 10.7-22.5%, with 99% of pituitary microadenomas, and it is estimated that the occurrence of pituitary microadenomas affects more than 7 million patients worldwide. Magnetic Resonance Imaging (MRI) is currently considered to be the primary method of pituitary imaging, and is performed by analysis and interpretation of pituitary MRI by clinicians or radiologists to diagnose whether a patient has pituitary adenoma.
However, in the process of implementing the invention, the inventor finds that the prior art has at least the following problems: since the pituitary gland is very small, the difference in anatomical structure between different individuals is large, and the diagnosis of pituitary adenoma mainly depends on the subjective judgment of a clinician or radiologist, which is a time-consuming and error-prone process, and the diagnosis accuracy is subject to the expertise and experience of the clinician. The current major diagnostic sensitivity is reported to be only 39-47%. Thus, the current diagnosis of nuclear magnetic resonance presents challenges and difficulties, making the diagnosis of pituitary adenomas time consuming and subjective.
Disclosure of Invention
The embodiment of the invention aims to provide a pituitary adenoma magnetic resonance image positioning diagnosis method and device based on artificial intelligence.
In order to achieve the above object, an embodiment of the present invention provides an artificial intelligence-based pituitary adenoma magnetic resonance image localization diagnosis method, including:
acquiring at least one pituitary magnetic resonance image of a person to be detected;
preprocessing each pituitary magnetic resonance image of the person to be detected, and stacking to obtain a corresponding magnetic resonance image to be detected;
inputting the magnetic resonance image to be detected into a pituitary adenoma recognition model obtained by pre-training so that the pituitary adenoma recognition model can calculate the magnetic resonance image to be detected;
and outputting the classification result of whether the magnetic resonance image to be detected contains pituitary adenoma or not.
As an improvement of the above scheme, the training method of the pituitary adenoma recognition model specifically comprises the following steps:
acquiring pituitary magnetic resonance images of a plurality of normal pituitary persons and pituitary adenomas patients; wherein, the same normal pituitary person or pituitary adenoma patient has at least one pituitary magnetic resonance image;
for each normal pituitary person or pituitary adenoma patient, preprocessing each corresponding pituitary magnetic resonance image, and stacking to obtain a corresponding first model training image; wherein the first model training image corresponds to a pre-labeled true result of whether or not pituitary adenoma is present;
initializing parameters of a first convolution neural network, and calculating the first model training image by using the first convolution neural network so as to output a prediction result of whether the first model training image contains pituitary adenoma;
calculating a first loss function according to the prediction result and the real result; wherein the first loss function is used to measure the difference between the predicted result and the true result;
and updating parameters of the first convolutional neural network by adopting a gradient descent optimization algorithm to reduce the first loss function until the first loss function tends to be minimized, so as to obtain a trained pituitary adenoma recognition model.
As an improvement of the above scheme, the calculating a first loss function according to the prediction result and the real result specifically includes:
calculating the first loss function according to the predicted result and the real result by the following calculation formula:
where p is ∈ {0, 1}, tableDisplaying the real result;
representing a prediction of the first convolutional neural network output; 1 indicates the presence of pituitary tumor, and 0 indicates the absence of pituitary tumor.
As an improvement of the above scheme, for each normal pituitary person or pituitary adenoma patient, preprocessing each corresponding pituitary mr image, and stacking to obtain a corresponding first model training image, specifically includes:
extracting a pituitary region in each pituitary magnetic resonance image corresponding to each normal pituitary person or pituitary adenoma patient to obtain a corresponding image block of each pituitary region;
and scaling each image block of the pituitary region to have the same resolution by adopting a bilinear interpolation method, stacking the image blocks, and performing pixel value normalization processing to obtain the first model training image.
As an improvement of the above scheme, for each normal pituitary person or pituitary adenoma patient, extracting the corresponding pituitary region in each pituitary magnetic resonance image to obtain the corresponding image block of each pituitary region, specifically including:
aiming at each normal pituitary person or pituitary adenoma patient, inputting each corresponding pituitary magnetic resonance image into a pituitary area positioning model obtained by pre-training for calculation so as to output a pituitary area in each pituitary magnetic resonance image;
and extracting a pituitary area in each pituitary magnetic resonance image to obtain each corresponding pituitary area image block.
As an improvement of the above scheme, the preprocessing each pituitary magnetic resonance image of the person to be measured and stacking the images to obtain a corresponding magnetic resonance image to be measured specifically includes:
extracting a pituitary region in each pituitary magnetic resonance image of the person to be detected to obtain a corresponding image block of each pituitary region;
and scaling each pituitary region image block of the person to be detected to have the same resolution by adopting a bilinear interpolation method, stacking the image blocks, and performing pixel value normalization processing to obtain a magnetic resonance image to be detected of the person to be detected.
As an improvement of the above scheme, the extracting a pituitary region in each pituitary magnetic resonance image of the person to be detected to obtain a corresponding image block of each pituitary region specifically includes:
inputting each pituitary magnetic resonance image of the person to be detected into a pituitary area positioning model obtained by pre-training for calculation so as to output a pituitary area in each pituitary magnetic resonance image of the person to be detected;
and extracting a pituitary region in each pituitary magnetic resonance image of the person to be detected to obtain each pituitary region image block corresponding to the person to be detected.
As an improvement of the above scheme, the training method of the pituitary region localization model specifically comprises the following steps:
acquiring a plurality of pituitary magnetic resonance images as second model training images; wherein each second model training image corresponds to a pre-labeled real pituitary region;
initializing parameters of a second convolutional neural network, and calculating the second model training image by using the second convolutional neural network so as to output a predicted pituitary region corresponding to the second model training image;
calculating a second loss function from the predicted pituitary region and the true pituitary region;
and updating the parameters of the second convolutional neural network by adopting a gradient descent optimization algorithm to reduce the second loss function until the second loss function tends to be minimized, and obtaining a trained pituitary region positioning model.
As an improvement of the above scheme, the calculating the second model training image by using the second convolutional neural network to output the predicted pituitary region corresponding to the second model training image specifically includes:
extracting feature maps of the second model training images to obtain a plurality of pyramid feature maps;
extracting candidate frames of each pyramid feature map to obtain a candidate frame set;
and eliminating the candidate frame excessively overlapped in the candidate frame set to obtain a target candidate frame serving as a predicted pituitary area.
The embodiment of the invention also provides an artificial intelligence-based pituitary adenoma magnetic resonance image positioning and diagnosing device, which comprises:
the image acquisition module is used for acquiring at least one pituitary magnetic resonance image of a person to be detected;
the preprocessing module is used for preprocessing each pituitary magnetic resonance image of the person to be detected and stacking the images to obtain a corresponding magnetic resonance image to be detected;
the image input module is used for inputting the magnetic resonance image to be detected into a pituitary adenoma recognition model obtained by pre-training so as to enable the pituitary adenoma recognition model to calculate the magnetic resonance image to be detected;
and the result output module is used for outputting the classification result of whether the magnetic resonance image to be detected contains pituitary adenomas.
Compared with the prior art, the pituitary adenoma magnetic resonance image positioning diagnosis method and device based on artificial intelligence disclosed by the invention have the advantages that a plurality of pituitary magnetic resonance images of a person to be detected are obtained, each pituitary magnetic resonance image is preprocessed, and the corresponding magnetic resonance images to be detected are obtained in a laminated mode. And then, inputting the magnetic resonance image to be detected into a pituitary adenoma recognition model obtained by pre-training, so that the pituitary adenoma recognition model can calculate the magnetic resonance image to be detected. And finally, outputting the classification result whether the magnetic resonance image to be detected contains pituitary adenoma. The pituitary magnetic resonance image of the person to be detected is calculated and analyzed by using the pre-constructed and trained pituitary adenoma recognition model, so as to recognize whether the pituitary magnetic resonance image of the person to be detected contains effective information of pituitary adenoma, thereby solving the problems of low efficiency and low accuracy caused by manual interpretation and labeling by doctors in the prior art, and effectively improving the efficiency and the precision of recognizing whether the pituitary adenoma is contained in the pituitary magnetic resonance image.
Drawings
FIG. 1 is a schematic flow chart illustrating the steps of an artificial intelligence-based magnetic resonance image localization diagnosis method for pituitary adenomas according to an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating the steps of a training method for a pituitary adenoma recognition model according to a second embodiment of the present invention;
FIG. 3 is a schematic diagram illustrating the steps of a method for training a pituitary region localization model according to a third embodiment of the present invention;
fig. 4 is a schematic structural diagram of an artificial intelligence-based magnetic resonance image positioning and diagnosing apparatus for pituitary adenoma according to a fourth embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a schematic flow chart of steps of an artificial intelligence-based magnetic resonance image localization diagnosis method for pituitary adenomas according to an embodiment of the present invention. The pituitary adenoma magnetic resonance image positioning diagnosis method based on artificial intelligence provided by the embodiment of the invention comprises the following steps of S11-S14:
and S11, acquiring at least one pituitary magnetic resonance image of the person to be detected.
And S12, preprocessing each pituitary magnetic resonance image of the person to be detected, and laminating to obtain a corresponding magnetic resonance image to be detected.
And S13, inputting the magnetic resonance image to be detected into a pituitary adenoma recognition model obtained by pre-training, so that the pituitary adenoma recognition model can calculate the magnetic resonance image to be detected.
And S14, outputting the classification result whether the magnetic resonance image to be detected contains pituitary adenoma.
In the embodiment of the invention, the person to be detected obtains intracranial MRI scanning data through MRI nuclear magnetic resonance examination. In order to ensure that whether the pituitary magnetic resonance image of the person to be detected contains pituitary adenoma or not is effectively identified, at least one MRI pituitary image which can clearly observe the pituitary is selected according to the intracranial MRI scanning data of the person to be detected, namely, the pituitary magnetic resonance image of the person to be detected is obtained.
Specifically, the pituitary magnetic resonance image of the person to be measured is a coronal pituitary magnetic resonance image, which is a layer of scan image with a thickness of 1mm, and the number of the pituitary magnetic resonance images is preferably five. By adopting the technical means of the embodiment of the invention, the validity and comprehensiveness of the pituitary magnetic resonance image of the person to be detected can be ensured, so that the identification precision of the pituitary adenoma is further improved.
And further, preprocessing five pituitary magnetic resonance images of the person to be detected to obtain corresponding magnetic resonance images to be detected, and inputting the magnetic resonance images to be detected into the pituitary adenoma recognition model for calculation.
Specifically, step S12 includes:
and S121, extracting a pituitary area in each pituitary magnetic resonance image of the person to be detected to obtain a corresponding image block of each pituitary area.
And S122, scaling each image block of the pituitary region of the person to be detected to have the same resolution by adopting a bilinear interpolation method, stacking the image blocks, and performing pixel value normalization processing to obtain a magnetic resonance image to be detected of the person to be detected.
In the embodiment of the invention, for each pituitary magnetic resonance image of the person to be detected, an invalid slice not containing the brain is removed by a global threshold segmentation method, and then the head irregular posture of the person to be detected is adjusted by front combination-rear combination correction when the pituitary magnetic resonance image is obtained; then, carrying out skull stripping and cerebellum removal on the pituitary magnetic resonance image so as to obtain a complete and single brain tissue; finally, all extracted brain tissue images are normalized to a uniform sample space by adjusting the spatial resolution, correcting intensity inhomogeneities using the N3 algorithm, and resampling using trilinear interpolation to eliminate differences between pituitary mr brain images obtained with different imaging devices.
Preferably, the spatial resolution of the pituitary mr images is adjusted to 1 × 1mm, intensity inhomogeneities are corrected using the N3 algorithm and resampling is performed to 128 × 128 using a trilinear interpolation method, thereby normalizing all processed pituitary mr images to a uniform sample space.
Furthermore, for the five pituitary magnetic resonance images of the person to be detected, corresponding pituitary regions need to be extracted, so as to improve the pertinence and accuracy of the pituitary adenoma.
The extraction method of the pituitary region can be obtained by manual labeling of doctors, or can be used for processing the pituitary magnetic resonance images by adopting a preset image processing algorithm to obtain the pituitary region in each pituitary magnetic resonance image, and further extracting and obtaining the image block of the pituitary region, without affecting the beneficial effects obtained by the invention.
In a preferred embodiment, step S121 is specifically executed by:
s1211, inputting each pituitary magnetic resonance image of the person to be detected into a pre-trained pituitary region positioning model for calculation, so as to output a pituitary region in each pituitary magnetic resonance image of the person to be detected;
and S1212, extracting a pituitary region in each pituitary magnetic resonance image of the person to be detected, and obtaining each pituitary region image block corresponding to the person to be detected.
In the embodiment of the invention, five pituitary magnetic resonance images of the person to be detected are input into the pituitary area positioning model for calculation through a pre-trained pituitary area positioning model, so that the corresponding pituitary area is output, and the corresponding five pituitary area image blocks are extracted and obtained.
Furthermore, each pituitary region image block of the person to be measured is scaled by bilinear interpolation, so that the five pituitary region image blocks all have the same resolution, namely 256 × 256. Then, the five pituitary image blocks are stacked, and the pixel value is normalized to [ -1,1], so as to obtain the magnetic resonance image to be measured, the size of which is 5 × 256 × 256.
Inputting the magnetic resonance image to be detected into a pre-trained pituitary adenoma recognition model for calculation, and outputting a classification result of whether the magnetic resonance image to be detected contains pituitary adenoma
It is possible to identify whether the test person has a pituitary adenoma, wherein,
1 indicates the presence of pituitary tumor, and 0 indicates the absence of pituitary tumor.
The embodiment of the invention provides an artificial intelligence-based pituitary adenoma magnetic resonance image positioning diagnosis method, which comprises the steps of obtaining a plurality of pituitary magnetic resonance images of a person to be detected, preprocessing each pituitary magnetic resonance image, and laminating to obtain a corresponding magnetic resonance image to be detected. And then, inputting the magnetic resonance image to be detected into a pituitary adenoma recognition model obtained by pre-training, so that the pituitary adenoma recognition model can calculate the magnetic resonance image to be detected. And finally, outputting the classification result whether the magnetic resonance image to be detected contains pituitary adenoma. The pituitary magnetic resonance image of the person to be detected is calculated and analyzed by using the pre-constructed and trained pituitary adenoma recognition model, so as to recognize whether the pituitary magnetic resonance image of the person to be detected contains effective information of pituitary adenoma, thereby solving the problems of low efficiency and low accuracy caused by manual interpretation and labeling by doctors in the prior art, and effectively improving the efficiency and the precision of recognizing whether the pituitary adenoma is contained in the pituitary magnetic resonance image.
Fig. 2 is a schematic step diagram of a training method for a pituitary adenoma recognition model according to a second embodiment of the present invention. The training method of the pituitary adenoma recognition model is performed through steps S21 to S25:
s21, acquiring pituitary magnetic resonance images of a plurality of normal pituitary persons and pituitary adenomas patients; wherein, the same normal pituitary person or pituitary adenoma patient has at least one pituitary magnetic resonance image;
s22, aiming at each normal pituitary person or pituitary adenoma patient, preprocessing each corresponding pituitary magnetic resonance image, and stacking to obtain a corresponding first model training image; wherein the first model training image corresponds to a pre-labeled true result of whether or not pituitary adenoma is present;
s23, initializing parameters of a first convolution neural network, and calculating the first model training image by using the first convolution neural network to output a prediction result of whether the first model training image contains pituitary adenoma;
s24, calculating a first loss function according to the prediction result and the real result; wherein the first loss function is used to measure the difference between the predicted result and the true result;
and S25, updating the parameters of the first convolution neural network by adopting a gradient descent optimization algorithm to reduce the first loss function until the first loss function tends to be minimized, and obtaining the trained pituitary adenoma recognition model.
In embodiments of the invention, intracranial MRI scan data is obtained for multiple normal pituitary persons and multiple pituitary adenomas. In order to ensure the effectiveness and accuracy of the model training process, for each normal pituitary person or pituitary adenoma patient, at least one pituitary magnetic resonance image in which the pituitary can be clearly observed needs to be selected from the corresponding MRI scan data. The pituitary magnetic resonance image is a coronal pituitary magnetic resonance image slice which is a layer of scanned image with the thickness of 1 mm.
Preferably, the number of pituitary magnetic resonance images per normal pituitary person or patient with pituitary adenoma is preferably five. Selecting 1000 pituitary magnetic resonance images of normal persons and 1000 pituitary magnetic resonance images of pituitary adenomas patients prestored in a database, namely 5000 pituitary magnetic resonance images of normal persons, 5000 pituitary magnetic resonance images of pituitary adenomas patients and 10000 pituitary magnetic resonance images in total, as training samples of the pituitary adenoma recognition model. For the same person (normal pituitary person or pituitary adenoma patient), the five pituitary magnetic resonance images thereof correspond to the real result of whether the pituitary adenoma exists, that is, the real result of the five pituitary magnetic resonance images of the normal pituitary person corresponds to the fact that the pituitary adenoma does not exist, and the real result of the five pituitary magnetic resonance images of the pituitary adenoma patient corresponds to the fact that the pituitary adenoma exists. The real result is pre-labeled.
Preferably, 5000 pituitary magnetic resonance images corresponding to normal pituitary persons or pituitary adenomas patients are divided into a model training set model test set according to a ratio of 9: 1. The model training set is used as a learning and training sample of the pituitary adenoma recognition model; and the model test set is used for testing the actual application environment after the primary training of the pituitary adenoma recognition model is finished.
Further, five pituitary magnetic resonance images corresponding to each normal pituitary person or pituitary adenoma patient are preprocessed to obtain a first model training image corresponding to each normal pituitary person or pituitary adenoma patient. And inputting the convolutional neural network for learning and training to obtain a pituitary adenoma recognition model.
Specifically, step S22 includes:
s221, aiming at each normal pituitary person or pituitary adenoma patient, extracting a corresponding pituitary area in each pituitary magnetic resonance image to obtain a corresponding image block of each pituitary area;
s222, scaling each image block of the pituitary region to have the same resolution by adopting a bilinear interpolation method, stacking the image blocks, and performing pixel value normalization processing to obtain the first model training image.
In the embodiment of the present invention, for the same person (normal pituitary person or pituitary adenoma patient), the corresponding image blocks of the pituitary region are extracted from the corresponding five pituitary magnetic resonance images, and then scaled by bilinear interpolation, so that the image blocks of the five pituitary regions all have the same resolution, i.e., 256 × 256. These five pituitary image blocks are then stacked and normalized to pixel values of [ -1,1], resulting in the first model training image, which is 5 × 256 × 256 in size. Through the steps, a plurality of first model training images can be obtained.
In the embodiment of the present invention, the extraction method of the pituitary region may be obtained by manual labeling performed by a doctor expert, or the preset image processing algorithm may be adopted to process the pituitary magnetic resonance image to obtain the pituitary region in each pituitary magnetic resonance image, and further extract and obtain the image block of the pituitary region, without affecting the beneficial effects obtained by the present invention.
In a preferred embodiment, step S221 is specifically executed by the following steps:
s1211, aiming at each normal pituitary person or pituitary adenoma patient, inputting each corresponding pituitary magnetic resonance image into a pituitary area positioning model obtained through pre-training for calculation, and outputting a pituitary area in each pituitary magnetic resonance image;
and S1212, extracting a pituitary region in each pituitary magnetic resonance image to obtain each corresponding pituitary region image block.
In the embodiment of the invention, five pituitary magnetic resonance images corresponding to each normal pituitary person or pituitary adenoma patient are input into the pituitary area positioning model for calculation through a pre-trained pituitary area positioning model, so that the corresponding pituitary areas are output, and the corresponding five pituitary area image blocks are extracted and obtained.
Further, after obtaining a first model training image of each of the normal pituitary persons or pituitary adenomas, the first model training image is calculated by using a first convolutional neural network which is initialized by parameters in advance. The first convolutional neural network comprises a plurality of hidden layers, and features are automatically learned from the first model training image so as to output a prediction result of whether the first model training image comprises pituitary adenoma.
Specifically, the parameters of the convolutional neural network are random before adjustment, the parameters of the first convolutional neural network are initialized, then the first model training image is input into the convolutional neural network, the forward step is executed, and the corresponding output probability of whether pituitary adenoma is suffered is calculated
If it is
The prediction result of the first model training image is that the pituitary tumor is suffered, if so, the pituitary tumor is not suffered
It is considered not to have pituitary tumors.
The forward step includes: and performing convolution, Relu activation, maximum pooling, global average pooling and the like on the first model training image, and finally performing full-link layer and softmax layer processing to obtain the output probability and obtain a prediction result.
In one embodiment, the input first model training image is first subjected to convolution operation with step size 2 by using 64 convolution kernels with size 7 × 7, the operation result is subjected to relu activation function, and then maximum pooling downsampling is performed by using kernels with size 3 × 3 to obtain an initial feature map f0The size is 64 × 64 × 64. Next, for the initial feature map f0Processing with several adjacent convolution, activation layers to obtain a feature map f1The size is 256 × 64 × 64; next, for the feature map f1Using a plurality ofProcessing adjacent convolution and activation layers, and finally obtaining a characteristic diagram f by 2-time down-sampling of a pooling layer2The size is 512 multiplied by 32; in turn, similarly by the characteristic diagram f1Obtaining a feature map f2Can be passed through the feature map f2Obtaining a feature map f3The size of which is 1024 × 16 × 16, and a feature map f is obtained in the same manner4The size is 2048 × 8 × 8. Finally, for the obtained feature map f4Obtaining a feature map f by using the global average pooling layer processinggapThe size is 2048 × 1 × 1.
In the obtained characteristic diagram fgapThen, the output probability is obtained through the processing of a full connection layer and a softmax layer
Will output probability
A normalization process is performed so that
Thereby obtaining the prediction result; wherein, if
Indicating the presence of pituitary tumor, if
Indicating that it does not suffer from pituitary tumors.
Calculating a first loss function according to the predicted result of the first training model and the real result pre-labeled by the first training model
To measure the difference between the predicted and true results:
wherein p belongs to {0, 1}, and represents a real result pre-labeled by the first training model;
representing a prediction of the first convolutional neural network output; 1 indicates the presence of pituitary tumor, and 0 indicates the absence of pituitary tumor.
And then updating the parameters of the first convolutional neural network by adopting a gradient descent optimization algorithm, performing recalculation on the first model training image by using the updated first convolutional neural network to obtain a new prediction result, and further calculating a new first loss function according to the new prediction result. And updating the parameters of the convolutional neural network by adopting a gradient descent optimization algorithm, so that the first loss function can be continuously reduced, and the difference between the predicted result and the real result is reduced. And continuously performing iterative training until the value of the first loss function tends to be minimized, namely after the parameters are adjusted for many times, the first convolution neural network training is finished when the first loss function cannot be reduced any more, so that the trained pituitary adenoma recognition model is obtained.
Preferably, after the pituitary adenoma recognition model is trained, pituitary magnetic resonance images in a model test set are obtained for testing, and the accuracy of the pituitary adenoma recognition model is verified, so that the recognition accuracy of the pituitary adenoma recognition model before being put into practical application is ensured.
The second embodiment of the invention provides a training method of a pituitary adenoma recognition model, which comprises the steps of firstly obtaining pituitary magnetic resonance images of a plurality of normal pituitary persons and pituitary adenoma patients as training image samples of the recognition model. Initializing parameters of the first convolution neural network by random numbers, inputting a first model training image, executing a forward step, and calculating the output probability of whether two categories of pituitary adenomas exist or not to obtain a prediction result. Then, a first loss function between the prediction result and the real result is calculated, a back propagation algorithm calculates the gradient of the first loss function relative to all parameter weights, and the value of the parameter of the first convolution neural network is updated by a gradient descent method so as to minimize the first loss function, thereby completing the training of the pituitary adenoma recognition model. According to the embodiment of the invention, the pituitary adenoma recognition model is established by adopting the convolutional neural network, accurate and effective characteristics are automatically acquired from the pituitary magnetic resonance image for learning, the precision and the generalization capability of the pituitary adenoma recognition model are improved, and the efficiency and the precision for recognizing whether the pituitary adenoma exists in the pituitary magnetic resonance image in practical application are effectively improved.
Fig. 3 is a schematic step diagram of a method for training a pituitary region location model according to a third embodiment of the present invention. The pituitary region localization model provided by the third embodiment of the invention is suitable for the artificial intelligence-based pituitary adenoma magnetic resonance image localization diagnosis process in the first embodiment, and is also suitable for the training process of the pituitary adenoma recognition model in the second embodiment.
The training method of the pituitary region localization model is performed through steps S31 to S34:
s31, acquiring a plurality of pituitary magnetic resonance images as second model training images; wherein each second model training image corresponds to a pre-labeled real pituitary region;
s32, initializing parameters of a second convolutional neural network, and calculating the second model training image by using the second convolutional neural network to output a predicted pituitary region corresponding to the second model training image;
s33, calculating a second loss function according to the predicted pituitary area and the real pituitary area; wherein the second loss function is used to measure the difference between the predicted pituitary region and the true pituitary region;
and S34, updating the parameters of the second convolutional neural network by adopting a gradient descent optimization algorithm to reduce the second loss function, and obtaining a trained pituitary region positioning model when the second loss function tends to be minimized.
In the embodiment of the invention, a plurality of pituitary magnetic resonance images are acquired as the second model training image for being used as the training sample of the pituitary region positioning model. Wherein, each pituitary magnetic resonance image corresponds to a real pituitary area which is marked in advance.
Preferably, dividing the acquired pituitary magnetic resonance images into a model training set and a model testing set according to a preset proportion, wherein the model training set is used as a learning and training sample of a pituitary region positioning model; and the model test set is used for testing the actual application environment after the primary training of the pituitary region positioning model is finished.
Preferably, each pituitary magnetic resonance image is an MRI slice with a size of H × W, and a preprocessing operation is further performed when obtaining the plurality of pituitary magnetic resonance images. Specifically, each pituitary magnetic resonance image is enlarged or reduced to 256 × 256 uniform size by a bilinear interpolation algorithm, then the pixel value of each pituitary magnetic resonance image is normalized to the range of [ -1,1], and finally a second model training image for inputting a second convolutional neural network for training is obtained.
And initializing the parameters of the second convolutional neural network, inputting the second model training image into the second convolutional neural network for calculation, and outputting the predicted pituitary region of the pituitary magnetic resonance image.
As a preferred embodiment, step S32 specifically includes:
s321, extracting feature maps of the second model training images to obtain a plurality of pyramid feature maps;
and inputting the second model training image into the second convolutional neural network, and extracting a plurality of feature maps with different resolutions of the second model training image. And then, fusing the obtained feature maps with different resolutions through a feature pyramid network to obtain a plurality of pyramid feature maps. As an example, 4 feature maps of different resolutions of the second model training image are obtained by a convolutional neural network, namely { f }1,f2,f3,f4}; feature map f by feature pyramid network1,f2,f3,f4Performing fusion to obtain 4 pyramid feature maps, i.e.
Specifically, the second convolutional neural network performs convolutional operation with step length of 2 on the input second model training image by using 64 convolutional kernels with size of 7 × 7, the operation result passes through relu activation function, and then maximum pooling downsampling is performed by using kernels with size of 3 × 3 to obtain an initial feature map f0The size is 64 × 64 × 64. For the initial feature map f0Processing with several adjacent convolution, activation layers to obtain a feature map f1The size is 256 × 64 × 64. For the feature map f1Processing by using a plurality of adjacent convolution and activation layers, and finally obtaining a characteristic graph f by 2-time down-sampling of a pooling layer2The size is 512 × 32 × 32. In turn, similarly by the characteristic diagram f1Obtaining a feature map f2Can be passed through the feature map f2Obtaining a feature map f3The size of which is 1024 × 16 × 16, and a feature map f is obtained4The size is 2048 × 8 × 8. Finally, feature maps { f } of 4 different resolutions of the image can be obtained1,f2,f3,f4}。
And fusing the four feature maps obtained above through a feature pyramid network. For the feature map f4Obtaining a pyramid feature map with 256 channels by using convolution kernel processing with the size of 1 multiplied by 1 and through relu activation function
The size is 256 × 8 × 8. To obtain f3Corresponding pyramid feature map
Firstly, the pyramid feature map of the upper layer is aligned
Up-sampling by a factor of 2, and then summing with a feature map f of the activation layer by convolution3Point-by-point addition is carried out to finally obtain a pyramid feature map with the size of 256 multiplied by 16
Similarly obtained pyramid feature map
In a manner of obtaining pyramid feature maps in sequence
The size is 256 × 32 × 32; and a pyramid feature map
The size is 256 × 64 × 64. Finally, 4 pyramid feature maps can be obtained
It should be noted that the size n × H of the feature map mentioned in the embodiment of the present inventioni×WiWhere n denotes the number of features of the feature map, i.e. each position on the feature map has n values, Hi、WiIndicating the height and width of the feature map.
And S322, extracting candidate frames of each pyramid feature map to obtain a candidate frame set.
In particular, for each pyramid profile f obtained as described abovei p(256×Hi×Wi) Firstly, obtaining candidate frame classification chart c through two branch networks (namely two layers of convolution activation layers) which do not share weight respectivelyi(2k×Hi×Wi) And candidate frame regression graph ri(4k×Hi×Wi)。
In the candidate frame classification chart ciIn (2 k), the candidate frame classification diagram ciHas 2k values per position, k representing the number of anchor boxes in each position, an anchor box being a box artificially set with a fixed size and position. Each anchor box includes two values that indicate whether a target is present in the anchor box. By way of example, each anchor frame includes two values, x1 and x2, when x1<At x2, considerA probability greater than 0.5 indicates the presence of a target in the anchor box, otherwise indicates the absence of a target. Regression of the graph r in the candidate boxiIn (4 k), the box regression candidate graph riEach anchor frame comprising 4 values, regressing the graph r for said candidate frameiThe 4 regression parameters of (1) are respectively the offset value, height and width.
Further, obtaining the candidate frame classification map ciAnd candidate frame regression graph riThen, the map c is classified in the candidate frameiThe target occurrence probability in the anchor frame of each position is traversed, and a candidate frame classification chart c is screened outiThere is an anchor frame for the target. Then, the anchor frame is mapped to the regression graph r of the candidate frameiTo obtain a regression graph riThereby obtaining a candidate frame (x)1,y1,h1,w1,p1) Wherein (x)1,y1) Indicates the center position of the candidate frame, h1And w1Height and width of the candidate box, p1Representing the candidate box confidence. Classifying the candidate frames corresponding to the same pyramid feature mapiAnd candidate frame regression graph riSeveral candidate boxes may be obtained.
For each pyramid feature map fi pPerforming the candidate frame extraction operation to obtain all candidate frames, and obtaining the candidate frame set { (x)j,yj,hj,wj,pj)}。
And S323, eliminating the candidate frame excessively overlapped in the candidate frame set to obtain a target candidate frame serving as a predicted pituitary area.
In the embodiment of the present invention, in order to obtain a candidate frame containing the pituitary, the candidate frame needs to be screened. And (4) eliminating the excessively overlapped candidate frame through a non-maximum suppression algorithm to obtain a predicted pituitary area.
Specifically, the confidence p of each candidate box is determinediDescending order, taking out the first candidate frame with highest confidence from the sequence, and then eliminating the overlap ratio of the first candidate frame and the sequence which is more than a certain valueCandidate boxes for threshold to form a new sequence. Then, the second candidate frame with the highest confidence coefficient is taken out from the new sequence, and then the candidate frames with the coincidence degree with the second candidate frame larger than a certain threshold value in the new sequence are eliminated. By analogy, the first candidate frame and the second candidate frame … are finally formed into a candidate frame set, and one candidate frame closest to the center of the image is selected as the target candidate frame, that is, the predicted pituitary region.
Further, the volume layer in the pituitary area positioning model contains a large number of random parameters, and the model is required to be suitable for pituitary positioning tasks through a training process. Specifically, 2 anchor frames (having a size of 32 × 32, 64 × 64, and an aspect ratio of 0.5) are set for each position, that is, k is 2. Then, the output predicted pituitary region is encoded to the corresponding real classification map
And true regression plots
In (1). The real classification map
And true regression plots
Obtained in the process of labeling the real pituitary area in advance.
Using a second loss function
Measuring the pituitary region localization model output the difference between the predicted pituitary region and the true pituitary region:
wherein j denotes the jth anchor frame, and when the jth anchor frame contains the pituitary,
if not, then,
Δxij＝(x-xa)/wa，Δyij＝(y-ya)/ha，Δwij＝log(w/wa)，Δhij＝log(h/ha)；(xa,ya,wa,ha) And (x, y, w, h) represent the real parameters of the anchor frame and the predicted parameters of the model, respectively, (Δ x)ij,Δyij,Δwij,Δhij) Regression of the graph r from the candidate boxesiTo obtain the compound. CE (-) and L1Smooth (-) represent the cross-entropy function and L1 smoothing function, respectively.
And
the real candidate frame classification graph and the real candidate frame regression graph are respectively.
And after the second loss function is obtained through calculation, updating parameters of the second convolutional neural network by adopting a gradient descent optimization algorithm, performing calculation again on the second model training image by utilizing the updated second convolutional neural network to obtain a new predicted pituitary area, and calculating a new first loss function according to the new predicted pituitary area. And updating parameters of the convolutional neural network by adopting a gradient descent optimization algorithm, so that the loss function can be continuously reduced, and the difference between the predicted pituitary area and the real pituitary area is reduced. And (4) finishing training of the second convolutional neural network by continuously iterating and training until the value of the second loss function tends to be minimized, thereby obtaining a trained pituitary region positioning model.
Preferably, after the pituitary region positioning model is trained, acquiring a pituitary magnetic resonance image in a model test set for testing, and verifying the accuracy of the pituitary region positioning model, thereby ensuring the identification accuracy of the pituitary region positioning model before being put into practical application.
The third embodiment of the invention provides a training method of a pituitary region positioning model, which comprises the steps of firstly obtaining a plurality of pituitary magnetic resonance images marked with pituitary regions in advance as training image samples of the positioning model. And initializing parameters of the second convolutional neural network by using random numbers, inputting a second model training image, calculating and positioning the pituitary region, and outputting a predicted pituitary region. And then calculating a second loss function between the predicted pituitary region and the real pituitary region, calculating the gradient of the second loss function relative to all the parameter weights by using a back propagation algorithm, and updating the value of the parameter of the second convolutional neural network by using a gradient descent method so as to minimize the second loss function, thereby completing the training of the pituitary region positioning model. According to the embodiment of the invention, a pituitary region positioning model is established by adopting a convolutional neural network, accurate and effective characteristics are automatically acquired from a pituitary magnetic resonance image for learning, the precision and generalization capability of the pituitary region positioning model are improved, and the efficiency and precision of positioning the region of the pituitary in the pituitary magnetic resonance image in practical application are effectively improved.
Fig. 4 is a schematic structural diagram of an artificial intelligence-based magnetic resonance image positioning diagnosis device for pituitary adenoma according to a fourth embodiment of the present invention. The fourth embodiment of the present invention provides an artificial intelligence-based pituitary adenoma magnetic resonance image positioning and diagnosing apparatus 40, which comprises:
an image obtaining module 41, configured to obtain at least one pituitary magnetic resonance image of a person to be measured;
the preprocessing module 42 is configured to preprocess each pituitary magnetic resonance image of the person to be detected, and stack the images to obtain a corresponding magnetic resonance image to be detected;
the image input module 43 is configured to input the magnetic resonance image to be detected into a pre-trained pituitary adenoma recognition model, so that the pituitary adenoma recognition model calculates the magnetic resonance image to be detected;
and the result output module 44 is configured to output a classification result of whether the magnetic resonance image to be detected contains pituitary adenomas.
It should be noted that the pituitary adenoma magnetic resonance image localization diagnosis apparatus based on artificial intelligence provided by the embodiment of the present invention is used for executing all the process steps of the pituitary adenoma magnetic resonance image localization diagnosis method based on artificial intelligence of the above embodiment, and the working principles and beneficial effects of the two are in one-to-one correspondence, and thus are not described again.
The fourth embodiment of the invention provides an artificial intelligence-based pituitary adenoma magnetic resonance image positioning diagnosis device, which is used for obtaining a plurality of pituitary magnetic resonance images of a person to be detected, preprocessing each pituitary magnetic resonance image and laminating to obtain a corresponding magnetic resonance image to be detected. And then, inputting the magnetic resonance image to be detected into a pituitary adenoma recognition model obtained by pre-training, so that the pituitary adenoma recognition model can calculate the magnetic resonance image to be detected. And finally, outputting the classification result whether the magnetic resonance image to be detected contains pituitary adenoma. The pituitary magnetic resonance image of the person to be detected is calculated and analyzed by using the pre-constructed and trained pituitary adenoma recognition model, so as to recognize whether the pituitary magnetic resonance image of the person to be detected contains effective information of pituitary adenoma, thereby solving the problems of low efficiency and low accuracy caused by manual interpretation and labeling by doctors in the prior art, and effectively improving the efficiency and the precision of recognizing whether the pituitary adenoma is contained in the pituitary magnetic resonance image.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-only memory (ROM), a Random Access Memory (RAM), or the like.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention.
Claims (10)
1. An artificial intelligence-based pituitary adenoma magnetic resonance image positioning diagnosis method is characterized by comprising the following steps:
acquiring at least one pituitary magnetic resonance image of a person to be detected;
preprocessing each pituitary magnetic resonance image of the person to be detected, and stacking to obtain a corresponding magnetic resonance image to be detected;
inputting the magnetic resonance image to be detected into a pituitary adenoma recognition model obtained by pre-training so that the pituitary adenoma recognition model can calculate the magnetic resonance image to be detected;
and outputting the classification result of whether the magnetic resonance image to be detected contains pituitary adenoma or not.
2. The pituitary adenoma magnetic resonance image localization diagnosis method based on artificial intelligence as claimed in claim 1, wherein the training method of the pituitary adenoma recognition model is specifically as follows:
acquiring pituitary magnetic resonance images of a plurality of normal pituitary persons and pituitary adenomas patients; wherein, the same normal pituitary person or pituitary adenoma patient has at least one pituitary magnetic resonance image;
for each normal pituitary person or pituitary adenoma patient, preprocessing each corresponding pituitary magnetic resonance image, and stacking to obtain a corresponding first model training image; wherein the first model training image corresponds to a pre-labeled true result of whether or not pituitary adenoma is present;
initializing parameters of a first convolution neural network, and calculating the first model training image by using the first convolution neural network so as to output a prediction result of whether the first model training image contains pituitary adenoma;
calculating a first loss function according to the prediction result and the real result; wherein the first loss function is used to measure the difference between the predicted result and the true result;
and updating parameters of the first convolutional neural network by adopting a gradient descent optimization algorithm to reduce the first loss function until the first loss function tends to be minimized, so as to obtain a trained pituitary adenoma recognition model.
3. The method for diagnosing the location of a pituitary adenoma magnetic resonance image based on artificial intelligence as claimed in claim 2, wherein the calculating the first loss function according to the predicted result and the real result specifically comprises:
calculating the first loss function according to the predicted result and the real result by the following calculation formula:
wherein p ∈ {0, 1}, which represents the true result;
representing a prediction of the first convolutional neural network output; 1 indicates the presence of pituitary tumor, and 0 indicates the absence of pituitary tumor.
4. The artificial intelligence-based pituitary adenoma magnetic resonance image localization diagnosis method according to claim 2, wherein for each normal pituitary person or pituitary adenoma patient, each pituitary magnetic resonance image corresponding to the normal pituitary person or pituitary adenoma patient is preprocessed and stacked to obtain a corresponding first model training image, specifically comprising:
extracting a pituitary region in each pituitary magnetic resonance image corresponding to each normal pituitary person or pituitary adenoma patient to obtain a corresponding image block of each pituitary region;
and scaling each image block of the pituitary region to have the same resolution by adopting a bilinear interpolation method, stacking the image blocks, and performing pixel value normalization processing to obtain the first model training image.
5. The artificial intelligence-based pituitary adenoma magnetic resonance image location diagnosis method as claimed in claim 4, wherein the extracting of the pituitary region in each of the pituitary magnetic resonance images corresponding to each of the normal pituitary persons or pituitary adenomas patients to obtain the image blocks of each of the corresponding pituitary regions specifically comprises:
aiming at each normal pituitary person or pituitary adenoma patient, inputting each corresponding pituitary magnetic resonance image into a pituitary area positioning model obtained by pre-training for calculation so as to output a pituitary area in each pituitary magnetic resonance image;
and extracting a pituitary area in each pituitary magnetic resonance image to obtain each corresponding pituitary area image block.
6. The artificial intelligence-based pituitary adenoma magnetic resonance image localization diagnosis method according to claim 1, wherein the preprocessing is performed on each pituitary magnetic resonance image of the person to be tested, and the preprocessing is performed on each pituitary magnetic resonance image of the person to be tested and the images are stacked to obtain the corresponding magnetic resonance image to be tested, specifically comprising:
extracting a pituitary region in each pituitary magnetic resonance image of the person to be detected to obtain a corresponding image block of each pituitary region;
and scaling each pituitary region image block of the person to be detected to have the same resolution by adopting a bilinear interpolation method, stacking the image blocks, and performing pixel value normalization processing to obtain a magnetic resonance image to be detected of the person to be detected.
7. The artificial intelligence-based pituitary adenoma magnetic resonance image location diagnosis method according to claim 6, wherein the extracting of the pituitary region in each of the pituitary magnetic resonance images of the person to be tested to obtain the corresponding image block of each pituitary region specifically comprises:
inputting each pituitary magnetic resonance image of the person to be detected into a pituitary area positioning model obtained by pre-training for calculation so as to output a pituitary area in each pituitary magnetic resonance image of the person to be detected;
and extracting a pituitary region in each pituitary magnetic resonance image of the person to be detected to obtain each pituitary region image block corresponding to the person to be detected.
8. The artificial intelligence-based pituitary adenoma magnetic resonance image location diagnosis method according to claim 5 or 7, characterized in that the training method of the pituitary region location model is specifically as follows:
acquiring a plurality of pituitary magnetic resonance images as second model training images; wherein each second model training image corresponds to a pre-labeled real pituitary region;
initializing parameters of a second convolutional neural network, and calculating the second model training image by using the second convolutional neural network so as to output a predicted pituitary region corresponding to the second model training image;
calculating a second loss function from the predicted pituitary region and the true pituitary region;
and updating the parameters of the second convolutional neural network by adopting a gradient descent optimization algorithm to reduce the second loss function until the second loss function tends to be minimized, and obtaining a trained pituitary region positioning model.
9. The artificial intelligence-based pituitary adenoma magnetic resonance image location diagnosis method according to claim 8, wherein the calculating the second model training image by using the second convolutional neural network to output the predicted pituitary region corresponding to the second model training image specifically comprises:
extracting feature maps of the second model training images to obtain a plurality of pyramid feature maps;
extracting candidate frames of each pyramid feature map to obtain a candidate frame set;
and eliminating the candidate frame excessively overlapped in the candidate frame set to obtain a target candidate frame serving as a predicted pituitary area.
10. An artificial intelligence-based pituitary adenoma magnetic resonance image positioning diagnosis device is characterized by comprising:
the image acquisition module is used for acquiring at least one pituitary magnetic resonance image of a person to be detected;
the preprocessing module is used for preprocessing each pituitary magnetic resonance image of the person to be detected and stacking the images to obtain a corresponding magnetic resonance image to be detected;
the image input module is used for inputting the magnetic resonance image to be detected into a pituitary adenoma recognition model obtained by pre-training so as to enable the pituitary adenoma recognition model to calculate the magnetic resonance image to be detected;
and the result output module is used for outputting the classification result of whether the magnetic resonance image to be detected contains pituitary adenomas.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010907454.2A CN112164028A (en) | 2020-09-02 | 2020-09-02 | Pituitary adenoma magnetic resonance image positioning diagnosis method and device based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010907454.2A CN112164028A (en) | 2020-09-02 | 2020-09-02 | Pituitary adenoma magnetic resonance image positioning diagnosis method and device based on artificial intelligence |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112164028A true CN112164028A (en) | 2021-01-01 |
Family
ID=73857661
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010907454.2A Pending CN112164028A (en) | 2020-09-02 | 2020-09-02 | Pituitary adenoma magnetic resonance image positioning diagnosis method and device based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112164028A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113077456A (en) * | 2021-04-20 | 2021-07-06 | 北京大学 | Training method and device for constructing network model based on functional magnetic resonance imaging, computer equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110400302A (en) * | 2019-07-25 | 2019-11-01 | 杭州依图医疗技术有限公司 | The method and device of lesion information in a kind of determination, display breast image |
| CN110415234A (en) * | 2019-07-29 | 2019-11-05 | 北京航空航天大学 | Brain tumor dividing method based on multi-parameter magnetic resonance imaging |
| CN110738649A (en) * | 2019-10-14 | 2020-01-31 | 青岛大学 | training method of Faster RCNN network for automatic identification of stomach cancer enhanced CT images |
| WO2020118618A1 (en) * | 2018-12-13 | 2020-06-18 | 深圳先进技术研究院 | Mammary gland mass image recognition method and device |
| CN111401480A (en) * | 2020-04-27 | 2020-07-10 | 上海市同济医院 | Novel breast MRI (magnetic resonance imaging) automatic auxiliary diagnosis method based on fusion attention mechanism |
| CN111583204A (en) * | 2020-04-27 | 2020-08-25 | 天津大学 | Organ positioning method of two-dimensional sequence magnetic resonance image based on network model |
-
2020
- 2020-09-02 CN CN202010907454.2A patent/CN112164028A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020118618A1 (en) * | 2018-12-13 | 2020-06-18 | 深圳先进技术研究院 | Mammary gland mass image recognition method and device |
| CN110400302A (en) * | 2019-07-25 | 2019-11-01 | 杭州依图医疗技术有限公司 | The method and device of lesion information in a kind of determination, display breast image |
| CN110415234A (en) * | 2019-07-29 | 2019-11-05 | 北京航空航天大学 | Brain tumor dividing method based on multi-parameter magnetic resonance imaging |
| CN110738649A (en) * | 2019-10-14 | 2020-01-31 | 青岛大学 | training method of Faster RCNN network for automatic identification of stomach cancer enhanced CT images |
| CN111401480A (en) * | 2020-04-27 | 2020-07-10 | 上海市同济医院 | Novel breast MRI (magnetic resonance imaging) automatic auxiliary diagnosis method based on fusion attention mechanism |
| CN111583204A (en) * | 2020-04-27 | 2020-08-25 | 天津大学 | Organ positioning method of two-dimensional sequence magnetic resonance image based on network model |
Non-Patent Citations (2)
| Title |
|---|
| QING LING LI 等: "Development and Validation of a Deep Learning Algorithm to Automatic Detection of Pituitary Microadenoma From MRI", 《FRONTIERS IN MEDICINE》, 29 November 2021 (2021-11-29), pages 1 - 10 * |
| 李理: "目标检测", Retrieved from the Internet <URL:https://fancyerii.github.io/books/object-detection/> * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113077456A (en) * | 2021-04-20 | 2021-07-06 | 北京大学 | Training method and device for constructing network model based on functional magnetic resonance imaging, computer equipment and storage medium |
| CN113077456B (en) * | 2021-04-20 | 2022-01-04 | 北京大学 | Training method and device for constructing network model based on functional magnetic resonance imaging |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111798462B (en) | Automatic delineation method of nasopharyngeal carcinoma radiotherapy target area based on CT image | |
| CN111784671B (en) | Pathological image focus region detection method based on multi-scale deep learning | |
| CN110503630B (en) | Cerebral hemorrhage classifying, positioning and predicting method based on three-dimensional deep learning model | |
| CN110930416B (en) | MRI image prostate segmentation method based on U-shaped network | |
| WO2018120942A1 (en) | System and method for automatically detecting lesions in medical image by means of multi-model fusion | |
| CN109584209B (en) | Vascular wall plaque recognition apparatus, system, method, and storage medium | |
| CN111047572A (en) | Automatic spine positioning method in medical image based on Mask RCNN | |
| CN111612856B (en) | Retina neovascularization detection method and imaging method for color fundus image | |
| CN111932550A (en) | 3D ventricle nuclear magnetic resonance video segmentation system based on deep learning | |
| KR20230059799A (en) | A Connected Machine Learning Model Using Collaborative Training for Lesion Detection | |
| CN112365464A (en) | GAN-based medical image lesion area weak supervision positioning method | |
| CN115004222A (en) | Neural network processing of OCT data to generate predictions of geographic atrophy growth rate | |
| CN113012086A (en) | Cross-modal image synthesis method | |
| CN115526801A (en) | Automatic color homogenizing method and device for remote sensing image based on conditional antagonistic neural network | |
| CN111383759A (en) | Automatic pneumonia diagnosis system | |
| CN112164028A (en) | Pituitary adenoma magnetic resonance image positioning diagnosis method and device based on artificial intelligence | |
| CN111402244B (en) | Automatic classification method for standard fetal heart tangent planes | |
| JP2023517058A (en) | Automatic detection of tumors based on image processing | |
| CN115601535A (en) | Chest radiograph abnormal recognition domain self-adaption method and system combining Wasserstein distance and difference measurement | |
| CN116309806A (en) | CSAI-Grid RCNN-based thyroid ultrasound image region of interest positioning method | |
| CN116128854A (en) | Hip joint ultrasonic image quality assessment method based on convolutional neural network | |
| CN112233126B (en) | Windowing method and device for medical image | |
| CN114332278A (en) | OCTA image motion correction method based on deep learning | |
| CN112053342A (en) | Method and device for extracting and identifying pituitary magnetic resonance image based on artificial intelligence | |
| CN111640126A (en) | Artificial intelligence diagnosis auxiliary method based on medical image |
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 |