CN116777883B - A method and system for identifying brain tumors in brain magnetic resonance images - Google Patents
A method and system for identifying brain tumors in brain magnetic resonance images Download PDFInfo
- Publication number
- CN116777883B CN116777883B CN202310775752.4A CN202310775752A CN116777883B CN 116777883 B CN116777883 B CN 116777883B CN 202310775752 A CN202310775752 A CN 202310775752A CN 116777883 B CN116777883 B CN 116777883B
- Authority
- CN
- China
- Prior art keywords
- slice
- slices
- subsequence
- mri
- sequence
- 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.)
- Active
Links
Landscapes
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention provides a brain tumor identification method and a brain tumor identification system in a brain nuclear magnetic resonance image, which are characterized by acquiring MRI of a minimum brain tumor region in a training sample set, acquiring the thickness of a layer, the interval between layers and the number of slices of the minimum brain tumor in the MRI, calculating the distance L according to the thickness of the layer, the interval between layers and the number of slices, acquiring a slice sequence corresponding to an MRI file to be identified, acquiring the thickness ST, the interval between layers SG and the number of slices SN of the MRI to be identified, acquiring the depth D of a 3D image which is constructed and required to be input, acquiring a similarity sequence corresponding to the slice sequence, acquiring D slices from the slice sequence according to the similarity sequence, the distance L, the interval between layers ST and the interval between layers SG to form a processed MRI file to be identified, and taking the processed MRI file to be identified as the input of a trained 3DU-Net to obtain an identification result. The invention not only utilizes the spatial information contained in the MRI, but also reduces the data volume in the 3D identification, and improves the accuracy and the speed.
Description
Technical Field
The invention relates to the field of image recognition, in particular to a brain tumor recognition method and system in brain nuclear magnetic resonance images.
Background
The brain is a control center of a human body, supports thinking, actions and the like of people, is a source of human intelligence, is a highly precise organ, and can seriously influence the normal life of people. Brain tumor is a disease caused by abnormal generation of intracranial cells, has become one of diseases seriously harming human health, clinically, usually, a noninvasive technology such as nuclear magnetic resonance (MRI, magnetic Resonance Imaging) is adopted to acquire brain images, then a doctor determines a lesion part, and then a treatment scheme is determined according to the characteristics of the lesion part, for example, surgical treatment, chemical treatment, radiation treatment and the like are adopted, so that the survival chance and the survival time of a patient are improved.
The recognition of brain tumor mainly depends on the judgment of doctor, which has a great relation with the experience of doctor, and small brain tumor recognition is more difficult due to the influence of human vision. The brain tumor is segmented by an automatic segmentation method of brain MRI, the segmentation effect of the segmentation technology based on deep learning is better than that of the traditional segmentation method, but the brain MRI is a three-dimensional image, if a2D segmentation method is adopted, some important information is lost, but the data size is too large, the training and the recognition are very slow, and errors exceeding the memory of the GPU are frequently reported by adopting a 3D segmentation method.
Disclosure of Invention
In order to solve the above problems, in a first aspect, the present invention provides a brain tumor recognition method in brain nuclear magnetic resonance images, the method comprising the steps of:
Acquiring a training sample set, calculating to obtain the MRI of the minimal brain tumor region marked in all training samples, and acquiring the layer thickness ST ' of the MRI, the layer spacing SG ' of the MRI and the slice number SN ' of the minimal brain tumor in the MRI;
Analyzing the brain nuclear magnetic resonance MRI file to be identified to obtain a slice sequence, and acquiring the layer thickness ST, the layer spacing SG and the slice number SN of the brain nuclear magnetic resonance MRI to be identified, and acquiring the depth D of a 3D image which is constructed and is required to be input by 3D U-Net;
obtaining D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG to form a processed MRI file to be identified;
and taking the processed MRI file to be identified as the trained 3D U-Net input to obtain an identification result.
Preferably, the MRI file to be identified after the processing is formed by obtaining D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG is specifically:
calculating a slice number sn″ included in the distance L from the layer thickness ST and the layer spacing SG;
Splitting the slice sequence into D subsequences from front to back, wherein the slices in the subsequences are continuous, and recording the splitting mode if each subsequence simultaneously meets the following conditions:
① The sub-sequence comprises at least one slice,
② The number of slices in the sub-sequence is less than SN',
③ The average value of the sums of the similarity subsequences corresponding to the subsequences is smaller than a threshold value;
Determining a segmentation mode to be used from the recorded segmentation modes, segmenting a slice sequence by adopting the segmentation mode to be used to obtain a plurality of subsequences, determining slices corresponding to each subsequence, and arranging the slices corresponding to each subsequence in sequence to obtain the MRI file to be identified.
Preferably, the determining the partition mode to be used from the recorded partition modes specifically includes:
for each recorded segmentation mode, calculating the similarity of the corresponding segmented sub-sequences, then calculating the standard deviation of the similarity of all the sub-sequences corresponding to the segmentation mode, and taking the segmentation mode with the minimum standard deviation as the segmentation mode to be used.
Preferably, the determining the slice corresponding to each sub-sequence specifically includes:
if the subsequence has only one slice, taking the slice contained in the subsequence as the slice corresponding to the subsequence;
If the subsequence has a plurality of slices, judging whether the number of the slices is odd, if so, selecting the slice in the middle of the subsequence as the slice corresponding to the subsequence, and if not, respectively calculating the similarity between the two slices in the middle and the previous slice and the similarity between the two slices in the middle and the subsequent slice to obtain the weights of the two slices, and selecting the slice with the smallest weight from the two slices in the middle as the slice corresponding to the subsequence.
Preferably, the MRI file to be identified after the processing is formed by obtaining D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG is specifically:
calculating a slice number sn″ included in the distance L from the layer thickness ST and the layer spacing SG;
Obtaining D-1 slice pairs with minimum similarity according to the similarity sequence, obtaining serial numbers of slices in the slice sequence in the slice pairs, obtaining cutting points according to the serial numbers, further obtaining D subsequences, judging whether the number of the slices in the subsequences is smaller than or equal to the number of the slices SN '', and if yes, re-cutting the sequence into three subsequences according to the similarity of the slices in the subsequences, the subsequences formed by the previous subsequences and the next subsequences of the subsequences;
And when the number of the slices in all the subsequences is larger than the number of the slices, recording D subsequences, determining the slices corresponding to each subsequence, and arranging the slices corresponding to each subsequence in sequence to obtain the MRI file to be identified.
In another aspect, the present invention further provides a brain tumor identification system in brain nuclear magnetic resonance images, the system comprising:
The distance calculation module is used for acquiring a training sample set, calculating and obtaining MRI (magnetic resonance imaging) of the minimal brain tumor region marked in all training samples, and acquiring a layer thickness ST ' of the MRI, an interlayer spacing SG ' of the MRI and a slice number SN ' of the minimal brain tumor in the MRI;
The analysis module is used for analyzing the brain nuclear magnetic resonance MRI file to be identified to obtain a slice sequence, acquiring the layer thickness ST, the layer spacing SG and the slice number SN of the brain nuclear magnetic resonance MRI to be identified, and acquiring the depth D of a 3D image which is constructed and is required to be input by 3D U-Net;
the dimension reduction module is used for calculating the similarity of two adjacent slices to obtain a similarity sequence corresponding to the slice sequence, and acquiring D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG to form a processed MRI file to be identified;
And the identification module is used for taking the processed MRI file to be identified as the trained input of 3D U-Net to obtain an identification result.
Preferably, the MRI file to be identified after the processing is formed by obtaining D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG is specifically:
calculating a slice number sn″ included in the distance L from the layer thickness ST and the layer spacing SG;
Splitting the slice sequence into D subsequences from front to back, wherein the slices in the subsequences are continuous, and recording the splitting mode if each subsequence simultaneously meets the following conditions:
① The sub-sequence comprises at least one slice,
② The number of slices in the sub-sequence is less than SN',
③ The average value of the sums of the similarity subsequences corresponding to the subsequences is smaller than a threshold value;
Determining a segmentation mode to be used from the recorded segmentation modes, segmenting a slice sequence by adopting the segmentation mode to be used to obtain a plurality of subsequences, determining slices corresponding to each subsequence, and arranging the slices corresponding to each subsequence in sequence to obtain the MRI file to be identified.
Preferably, the determining the partition mode to be used from the recorded partition modes specifically includes:
for each recorded segmentation mode, calculating the similarity of the corresponding segmented sub-sequences, then calculating the standard deviation of the similarity of all the sub-sequences corresponding to the segmentation mode, and taking the segmentation mode with the minimum standard deviation as the segmentation mode to be used.
Preferably, the determining the slice corresponding to each sub-sequence specifically includes:
if the subsequence has only one slice, taking the slice contained in the subsequence as the slice corresponding to the subsequence;
If the subsequence has a plurality of slices, judging whether the number of the slices is odd, if so, selecting the slice in the middle of the subsequence as the slice corresponding to the subsequence, and if not, respectively calculating the similarity between the two slices in the middle and the previous slice and the similarity between the two slices in the middle and the subsequent slice to obtain the weights of the two slices, and selecting the slice with the smallest weight from the two slices in the middle as the slice corresponding to the subsequence.
Preferably, the MRI file to be identified after the processing is formed by obtaining D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG is specifically:
calculating a slice number sn″ included in the distance L from the layer thickness ST and the layer spacing SG;
Obtaining D-1 slice pairs with minimum similarity according to the similarity sequence, obtaining serial numbers of slices in the slice sequence in the slice pairs, obtaining cutting points according to the serial numbers, further obtaining D subsequences, judging whether the number of the slices in the subsequences is smaller than or equal to the number of the slices SN '', and if yes, re-cutting the sequence into three subsequences according to the similarity of the slices in the subsequences, the subsequences formed by the previous subsequences and the next subsequences of the subsequences;
And when the number of the slices in all the subsequences is larger than the number of the slices, recording D subsequences, determining the slices corresponding to each subsequence, and arranging the slices corresponding to each subsequence in sequence to obtain the MRI file to be identified.
Finally, the invention also provides a computer-readable storage medium having stored thereon a computer program, characterized in that the computer program, when being executed by a processor, implements a method as described above.
Aiming at the problems that in the prior art, in brain tumor identification of brain magnetic resonance images, the 2D identification or segmentation method cannot utilize spatial information, and the calculation amount of the 3D identification or segmentation method is large, the brain nuclear magnetic resonance MRI file is processed, specifically, the brain nuclear magnetic resonance MRI file to be identified is processed according to the similarity of two adjacent slices and the number of the minimum brain tumor occupied slices in a training sample, so that the number of slices in the MRI file is reduced, the calculation amount is reduced, the spatial information of MRI is reserved, and the identification effect is better.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a first embodiment;
FIG. 2 is a schematic illustration of layer spacing and layer thickness in MRI;
FIG. 3 is a slice sequence-similarity sequence correspondence graph;
Fig. 4 is a diagram showing a second embodiment.
Detailed Description
In this document, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises an element.
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The traditional screenshot mode is that a doctor manually intercepts pictures at key positions such as the pharyngeal portion, the laryngeal portion and the like in the examination process, and based on the screenshot mode, the invention provides a method and a system capable of automatically intercepting pictures at the key positions in the electronic nasopharynoscopy.
First embodiment
In a first aspect, the present invention provides a method for brain tumor identification in brain nuclear magnetic resonance images, as shown in fig. 1, the method comprising the steps of:
Step 1, acquiring a training sample set, calculating and obtaining MRI (magnetic resonance imaging) of the minimal brain tumor region marked in all training samples, and acquiring a layer thickness ST ' of the MRI, an interval SG ' of the MRI and a slice number SN ' of the minimal brain tumor in the MRI;
After labeling the training sample set, information such as labeling positions can be checked, and because different training sample layers (SLICE THICKNESS) and layer intervals (gaps) are different, the schematic diagrams of the layer thicknesses and the layer intervals are shown in fig. 2, and the same number of slices (slices) are different in space distance, if the slices extracted from the brain nuclear magnetic resonance MRI to be identified are too sparse, a small tumor area can possibly be unrecognized, and the invention firstly finds out the smallest brain tumor area in the sample training set, wherein the smallest brain tumor area is the brain tumor area with the shortest distance of the labeled brain tumor area in the vertical direction of the slices. And obtaining a minimum brain tumor area according to the brain tumor area marked in the sample, wherein the distance of the brain tumor area in the vertical slice direction is L. In one embodiment, L is calculated by SN '×st' + (SN '-1) ×sg'.
Step 2, analyzing the brain nuclear magnetic resonance MRI file to be identified to obtain a slice sequence, and acquiring the layer thickness ST, the layer spacing SG and the slice number SN of the brain nuclear magnetic resonance MRI to be identified, and acquiring the depth D of a 3D image which is constructed and is required to be input by 3D U-Net;
MRI consists of many slices, each slice being a picture, MRI being effectively a 3D image, with three dimensions of length, width and depth, the depth being the number of slices. And the MRI file analysis can obtain the related information of the MRI, such as the layer thickness, the layer spacing, the slice number and the like, and can obtain the slice sequence included in the MRI file. Because the input requirements of the 3D U-Net are the same, the processing and the segmentation can be uniformly carried out, and the length, the width and the depth of the constructed 3D image which is required to be input by the 3D U-Net are mainly acquired.
In the invention, the thickness, spacing and number of slices of the nuclear magnetic resonance MRI of the brain to be identified are not required to be the same, i.e. the MRI to be identified can come from different equipment and be obtained by scanning the brain with different parameters. The layer thickness, layer spacing, and number of slices of the training sample MRI are not required to be the same, since the same steps 2,3, and 4 will be performed on the training sample, except that the training sample is for training 3D U-Net. In addition, the length, width and depth of the training samples are not required to be the same, and the length, width and depth of the MRI to be identified are not required to be the same, because the 3D image is subjected to shape or reshape operation according to the input requirement of the built 3D U-Net.
Step 3, calculating the similarity of two adjacent slices to obtain a similarity sequence corresponding to the slice sequence, and acquiring D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG to form a processed MRI file to be identified;
in order to reduce the data to be processed by 3D U-Net, MRI needs to be compressed in depth direction as much as possible, and two or more adjacent slices with high similarity can be combined into one slice or one slice can be selected in the compression process, but too many consecutive slices cannot be combined for the accuracy of identification, which easily causes brain tumor segmentation failure. The invention obtains the MRI file to be identified after the composition treatment of D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG.
The similarity sequence corresponds to the slice sequence, and the elements in the similarity sequence are similarity values of the corresponding serial number slice and the subsequent slice in the slice sequence. Fig. 3 shows the correspondence between slice sequences and similarity sequences. According to the sequence numbers of the elements in the similarity sequence, two adjacent slices for calculating the similarity in the corresponding slice sequence can be found.
And step 4, taking the processed MRI file to be identified as the trained input of 3D U-Net to obtain an identification result.
After the MRI file to be identified passes through the steps 2 and 3, the number of contained slices is greatly reduced, so that the identification speed can be improved, and the spatial information of MRI is utilized, so that the identification result is more accurate.
In a specific embodiment, the method obtains the MRI file to be identified after the processing of the D slice compositions from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG, specifically:
calculating a slice number sn″ included in the distance L from the layer thickness ST and the layer spacing SG;
In order to ensure that the processed MRI to be identified contains brain tumor information, that is, in order to avoid missing possible tumors, the range of slice merging cannot be too wide, the distance between the merged slices must be ensured not to be greater than L, and the number sn″ of slices included in the distance L, specifically sn″ = (l+sg)/(st+sg) can be calculated according to the layer thickness ST and the layer spacing SG.
Splitting the sequence of slices from front to back into D sub-sequences, the slices in the sub-sequences being consecutive and the next slice of the last slice of the previous sub-sequence being the first slice of the next sub-sequence, recording the splitting pattern if each sub-sequence simultaneously satisfies the following conditions:
① The sub-sequence comprises at least one slice,
② The number of slices in the sub-sequence is less than SN',
③ The average value of the sums of the similarity subsequences corresponding to the subsequences is smaller than a threshold value;
Determining a segmentation mode to be used from the recorded segmentation modes, segmenting a slice sequence by adopting the segmentation mode to be used to obtain a plurality of subsequences, determining slices corresponding to each subsequence, and arranging the slices corresponding to each subsequence in sequence to obtain the MRI file to be identified.
For example, if the slice sequence includes 20 slices, and d=4, then one of the splitting modes is (1, 2,3, 4), (5, 6), (7, 8,9,10,11, 12), (13,14,15,16,17,18,19,20), and sn″=3, and the number of slices included in the second sub-sequence in this splitting mode is less than sn″, which is not satisfied. After the condition ①② is satisfied, the amount of computation may be reduced in determining whether ③ is satisfied. The calculation of condition ③ may be aided by a similarity sequence, which further reduces duplicate calculations.
The dividing modes are various, and in a specific embodiment, the dividing modes to be used are determined from the recorded dividing modes, specifically, the dividing modes are given by the principle that slices with high similarity are divided into a subsequence as much as possible:
for each recorded segmentation mode, calculating the similarity of the corresponding segmented sub-sequences, then calculating the standard deviation of the similarity of all the sub-sequences corresponding to the segmentation mode, and taking the segmentation mode with the minimum standard deviation as the segmentation mode to be used.
There are various ways of calculating the similarity of the subsequences, and in a more specific embodiment, a similarity subsequence corresponding to the subsequence is obtained, an average value of the similarity subsequences is calculated, and the average value is used as the similarity of the subsequence. It should be noted, however, that either the similarity sequence or the similarity subsequence will have one less element than the corresponding slice sequence. For example, a slice sub-sequence has 10 elements, then the sub-sequence corresponds to a similarity sub-sequence having 9 elements, since the last slice of the slice sub-sequence has no calculated similarity with the following slice not belonging to the slice sub-sequence. There are various ways to calculate the similarity of the pictures, and this will not be described in detail here.
After the segmentation is obtained, a plurality of slice sub-sequences are obtained, and the slice corresponding to each slice sub-sequence needs to be further determined, and in a specific embodiment, the determining the slice corresponding to each sub-sequence is specifically:
if the subsequence has only one slice, taking the slice contained in the subsequence as the slice corresponding to the subsequence;
If the subsequence has a plurality of slices, judging whether the number of the slices is odd, if so, selecting the slice in the middle of the subsequence as the slice corresponding to the subsequence, and if not, respectively calculating the similarity between the two slices in the middle and the previous slice and the similarity between the two slices in the middle and the subsequent slice to obtain the weights of the two slices, and selecting the slice with the smallest weight from the two slices in the middle as the slice corresponding to the subsequence.
In another embodiment, the determining the slice corresponding to each sub-sequence specifically includes obtaining the feature map as the slice corresponding to the sub-sequence by convolution or downsampling. The convolution kernel is related to the number of slices contained in the subsequence, and the maximum pooling in the depth direction is preferentially used for downsampling, wherein the maximum pooling in the depth direction refers to the value of a pixel point corresponding to the feature map obtained by the maximum pixel point in the depth direction.
In another embodiment, the MRI file to be identified after the processing of the D slice compositions is obtained from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG, specifically:
calculating a slice number sn″ included in the distance L from the layer thickness ST and the layer spacing SG;
Obtaining D-1 slice pairs with minimum similarity according to the similarity sequence, obtaining serial numbers of slices in the slice sequence in the slice pairs, obtaining cutting points according to the serial numbers, further obtaining D subsequences, judging whether the number of the slices in the subsequences is smaller than or equal to the number of the slices SN '', and if yes, re-cutting the sequence into three subsequences according to the similarity of the slices in the subsequences, the subsequences formed by the previous subsequences and the next subsequences of the subsequences;
And when the number of the slices in all the subsequences is larger than the number of the slices, recording D subsequences, determining the slices corresponding to each subsequence, and arranging the slices corresponding to each subsequence in sequence to obtain the MRI file to be identified.
Second embodiment
The invention also provides a brain tumor identification system in brain nuclear magnetic resonance images, which comprises the following modules:
The distance calculation module is used for acquiring a training sample set, calculating and obtaining MRI (magnetic resonance imaging) of the minimal brain tumor region marked in all training samples, and acquiring a layer thickness ST ' of the MRI, an interlayer spacing SG ' of the MRI and a slice number SN ' of the minimal brain tumor in the MRI;
The analysis module is used for analyzing the brain nuclear magnetic resonance MRI file to be identified to obtain a slice sequence, acquiring the layer thickness ST, the layer spacing SG and the slice number SN of the brain nuclear magnetic resonance MRI to be identified, and acquiring the depth D of a 3D image which is constructed and is required to be input by 3D U-Net;
the dimension reduction module is used for calculating the similarity of two adjacent slices to obtain a similarity sequence corresponding to the slice sequence, and acquiring D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG to form a processed MRI file to be identified;
And the identification module is used for taking the processed MRI file to be identified as the trained input of 3D U-Net to obtain an identification result.
Preferably, the MRI file to be identified after the processing is formed by obtaining D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG is specifically:
calculating a slice number sn″ included in the distance L from the layer thickness ST and the layer spacing SG;
Splitting the slice sequence into D subsequences from front to back, wherein the slices in the subsequences are continuous, and recording the splitting mode if each subsequence simultaneously meets the following conditions:
① The sub-sequence comprises at least one slice,
② The number of slices in the sub-sequence is less than SN',
③ The average value of the sums of the similarity subsequences corresponding to the subsequences is smaller than a threshold value;
Determining a segmentation mode to be used from the recorded segmentation modes, segmenting a slice sequence by adopting the segmentation mode to be used to obtain a plurality of subsequences, determining slices corresponding to each subsequence, and arranging the slices corresponding to each subsequence in sequence to obtain the MRI file to be identified.
Preferably, the determining the partition mode to be used from the recorded partition modes specifically includes:
for each recorded segmentation mode, calculating the similarity of the corresponding segmented sub-sequences, then calculating the standard deviation of the similarity of all the sub-sequences corresponding to the segmentation mode, and taking the segmentation mode with the minimum standard deviation as the segmentation mode to be used.
Preferably, the determining the slice corresponding to each sub-sequence specifically includes:
if the subsequence has only one slice, taking the slice contained in the subsequence as the slice corresponding to the subsequence;
If the subsequence has a plurality of slices, judging whether the number of the slices is odd, if so, selecting the slice in the middle of the subsequence as the slice corresponding to the subsequence, and if not, respectively calculating the similarity between the two slices in the middle and the previous slice and the similarity between the two slices in the middle and the subsequent slice to obtain the weights of the two slices, and selecting the slice with the smallest weight from the two slices in the middle as the slice corresponding to the subsequence.
Preferably, the MRI file to be identified after the processing is formed by obtaining D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG is specifically:
calculating a slice number sn″ included in the distance L from the layer thickness ST and the layer spacing SG;
Obtaining D-1 slice pairs with minimum similarity according to the similarity sequence, obtaining serial numbers of slices in the slice sequence in the slice pairs, obtaining cutting points according to the serial numbers, further obtaining D subsequences, judging whether the number of the slices in the subsequences is smaller than or equal to the number of the slices SN '', and if yes, re-cutting the sequence into three subsequences according to the similarity of the slices in the subsequences, the subsequences formed by the previous subsequences and the next subsequences of the subsequences;
And when the number of the slices in all the subsequences is larger than the number of the slices, recording D subsequences, determining the slices corresponding to each subsequence, and arranging the slices corresponding to each subsequence in sequence to obtain the MRI file to be identified.
Third embodiment
The present invention provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method as described in embodiment one of the present invention.
Fourth embodiment
The invention provides a computer device comprising at least a memory and a processor, the memory having stored thereon a computer program which, when executed by the processor, implements a method as described in the first embodiment of the invention.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by adding necessary general purpose hardware platforms, or may be implemented by a combination of hardware and software. Based on such understanding, the foregoing aspects, in essence and portions contributing to the art, may be embodied in the form of a computer program product, which may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
It should be noted that the above-mentioned embodiments are merely for illustrating the technical solution of the present invention, and not for limiting the same, and although the present invention has been described in detail with reference to the above-mentioned embodiments, it should be understood by those skilled in the art that the technical solution described in the above-mentioned embodiments may be modified or some technical features may be equivalently replaced, and these modifications or substitutions do not make the essence of the corresponding technical solution deviate from the spirit and scope of the technical solution of the embodiments of the present invention.
Claims (7)
1. A method for identifying brain tumors in brain nuclear magnetic resonance images, the method comprising the steps of:
acquiring a training sample set, calculating to obtain MRI (magnetic resonance imaging) of the minimal brain tumor region marked in all training samples, and acquiring a layer thickness ST ́ of the MRI, an interlayer spacing SG ́ of the MRI and a slice number SN ́ of the minimal brain tumor in the MRI, wherein a distance L is calculated according to the layer thickness ST ́, the interlayer spacing SG ́ and the slice number SN ́;
Analyzing the brain nuclear magnetic resonance MRI file to be identified to obtain a slice sequence, and acquiring the layer thickness ST, the layer spacing SG and the slice number SN of the brain nuclear magnetic resonance MRI to be identified, and acquiring the depth D of a 3D image which is constructed and is required to be input by 3D U-Net;
obtaining D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG to form a processed MRI file to be identified;
taking the processed MRI file to be identified as the trained 3D U-Net input to obtain an identification result;
the method comprises the steps of obtaining D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG to form a processed MRI file to be identified, wherein the MRI file to be identified specifically comprises the following steps:
calculating a slice number sn″ included in the distance L from the layer thickness ST and the layer spacing SG;
Splitting the slice sequence into D subsequences from front to back, wherein the slices in the subsequences are continuous, and recording the splitting mode if each subsequence simultaneously meets the following conditions:
① The sub-sequence comprises at least one slice,
② The number of slices in the sub-sequence is less than SN',
③ The average value of the sums of the similarity subsequences corresponding to the subsequences is smaller than a threshold value;
Determining a segmentation mode to be used from recorded segmentation modes, segmenting a slice sequence by adopting the segmentation mode to be used to obtain a plurality of subsequences, determining slices corresponding to each subsequence, and sequentially arranging the slices corresponding to each subsequence to obtain an MRI file to be identified, wherein determining the slices corresponding to each subsequence comprises selecting the slices positioned in the middle of the subsequence as the slices corresponding to the subsequence.
2. The method according to claim 1, wherein the determining the segmentation method to be used from the recorded segmentation methods, in particular, is:
for each recorded segmentation mode, calculating the similarity of the corresponding segmented sub-sequences, then calculating the standard deviation of the similarity of all the sub-sequences corresponding to the segmentation mode, and taking the segmentation mode with the minimum standard deviation as the segmentation mode to be used.
3. The method according to claim 1, wherein said determining the slice corresponding to each of said sub-sequences is in particular:
if the subsequence has only one slice, taking the slice contained in the subsequence as the slice corresponding to the subsequence;
If the subsequence has a plurality of slices, judging whether the number of the slices is odd, if so, selecting the slice in the middle of the subsequence as the slice corresponding to the subsequence, and if not, respectively calculating the similarity between the two slices in the middle and the previous slice and the similarity between the two slices in the middle and the subsequent slice to obtain the weights of the two slices, and selecting the slice with the smallest weight from the two slices in the middle as the slice corresponding to the subsequence.
4. A brain tumor recognition system in brain nuclear magnetic resonance images, the system comprising the following modules:
The distance calculation module is used for obtaining a training sample set, calculating and obtaining MRI (magnetic resonance imaging) of the minimal brain tumor region marked in all training samples, and obtaining a layer thickness ST ́ of the MRI, an interlayer spacing SG ́ of the MRI and a slice number SN ́ of the minimal brain tumor in the MRI;
The analysis module is used for analyzing the brain nuclear magnetic resonance MRI file to be identified to obtain a slice sequence, acquiring the layer thickness ST, the layer spacing SG and the slice number SN of the brain nuclear magnetic resonance MRI to be identified, and acquiring the depth D of a 3D image which is constructed and is required to be input by 3D U-Net;
the dimension reduction module is used for calculating the similarity of two adjacent slices to obtain a similarity sequence corresponding to the slice sequence, and acquiring D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG to form a processed MRI file to be identified;
The recognition module is used for taking the processed MRI file to be recognized as the trained input of 3D U-Net to obtain a recognition result;
the method comprises the steps of obtaining D slices from the slice sequence according to the similarity sequence, the distance L, the layer thickness ST and the layer spacing SG to form a processed MRI file to be identified, wherein the MRI file to be identified specifically comprises the following steps:
calculating a slice number sn″ included in the distance L from the layer thickness ST and the layer spacing SG;
Splitting the slice sequence into D subsequences from front to back, wherein the slices in the subsequences are continuous, and recording the splitting mode if each subsequence simultaneously meets the following conditions:
① The sub-sequence comprises at least one slice,
② The number of slices in the sub-sequence is less than SN',
③ The average value of the sums of the similarity subsequences corresponding to the subsequences is smaller than a threshold value;
Determining a segmentation mode to be used from recorded segmentation modes, segmenting a slice sequence by adopting the segmentation mode to be used to obtain a plurality of subsequences, determining slices corresponding to each subsequence, and sequentially arranging the slices corresponding to each subsequence to obtain an MRI file to be identified, wherein determining the slices corresponding to each subsequence comprises selecting the slices positioned in the middle of the subsequence as the slices corresponding to the subsequence.
5. The system according to claim 4, wherein the determining the segmentation method to be used from the recorded segmentation methods is specifically:
for each recorded segmentation mode, calculating the similarity of the corresponding segmented sub-sequences, then calculating the standard deviation of the similarity of all the sub-sequences corresponding to the segmentation mode, and taking the segmentation mode with the minimum standard deviation as the segmentation mode to be used.
6. The system of claim 4, wherein the determining the slice corresponding to each of the subsequences is specifically:
if the subsequence has only one slice, taking the slice contained in the subsequence as the slice corresponding to the subsequence;
If the subsequence has a plurality of slices, judging whether the number of the slices is odd, if so, selecting the slice in the middle of the subsequence as the slice corresponding to the subsequence, and if not, respectively calculating the similarity between the two slices in the middle and the previous slice and the similarity between the two slices in the middle and the subsequent slice to obtain the weights of the two slices, and selecting the slice with the smallest weight from the two slices in the middle as the slice corresponding to the subsequence.
7. A computer readable storage medium having stored thereon a computer program, which when executed by a processor, implements the method according to any of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310775752.4A CN116777883B (en) | 2023-06-28 | 2023-06-28 | A method and system for identifying brain tumors in brain magnetic resonance images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310775752.4A CN116777883B (en) | 2023-06-28 | 2023-06-28 | A method and system for identifying brain tumors in brain magnetic resonance images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116777883A CN116777883A (en) | 2023-09-19 |
| CN116777883B true CN116777883B (en) | 2025-03-21 |
Family
ID=88013092
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310775752.4A Active CN116777883B (en) | 2023-06-28 | 2023-06-28 | A method and system for identifying brain tumors in brain magnetic resonance images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116777883B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112734737A (en) * | 2021-01-18 | 2021-04-30 | 天津大学 | Intelligent capture system for brain glioma key frames based on VGG twin network |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2002239268A1 (en) * | 2000-11-24 | 2002-06-11 | U-Systems, Inc. | Method and system for ultrasound breast cancer screening |
| US9652871B2 (en) * | 2015-01-28 | 2017-05-16 | Impac Medical Systems, Inc. | Three dimensional localization of a moving target for adaptive radiation therapy |
| US10535001B2 (en) * | 2017-11-06 | 2020-01-14 | International Business Machines Corporation | Reducing problem complexity when analyzing 3-D images |
-
2023
- 2023-06-28 CN CN202310775752.4A patent/CN116777883B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112734737A (en) * | 2021-01-18 | 2021-04-30 | 天津大学 | Intelligent capture system for brain glioma key frames based on VGG twin network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116777883A (en) | 2023-09-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tang et al. | Automatic pulmonary lobe segmentation using deep learning | |
| CN111476796B (en) | Semi-supervised coronary artery segmentation system and segmentation method combining multiple networks | |
| CN111602173B (en) | Brain tomography data analysis method | |
| WO2019200753A1 (en) | Lesion detection method, device, computer apparatus and storage medium | |
| US12381007B2 (en) | Pancreatic postoperative diabetes prediction system based on supervised deep subspace learning | |
| JPH10500321A (en) | Automated method and system for detecting lesions in medical computed tomography scans | |
| KR101571151B1 (en) | Method of artifacts elimination for computed tomography image | |
| CN117726633B (en) | Segmentation method and system of double-branch coronary artery images based on feature fusion | |
| CN111402218A (en) | Cerebral hemorrhage detection method and device | |
| CN115311193B (en) | Abnormal brain image segmentation method and system based on dual-attention mechanism | |
| CN114092450A (en) | Real-time image segmentation method, system and device based on gastroscopy video | |
| CN114664409A (en) | Recurrent nasopharyngeal carcinoma image omics data processing method, system, device and storage medium | |
| Irene et al. | Segmentation and approximation of blood volume in intracranial hemorrhage patients based on computed tomography scan images using deep learning method | |
| US20160260210A1 (en) | Method and system for determining the prognosis of a patient suffering from pulmonary embolism | |
| Amaludin et al. | Toward more accurate diagnosis of multiple sclerosis: automated lesion segmentation in brain magnetic resonance image using modified U‐net model | |
| ABOUDI et al. | A Hybrid Model for Ischemic Stroke Brain Segmentation from MRI Images using CBAM and ResNet50-UNet. | |
| CN115116625B (en) | Pancreatic cancer lymph node metastasis stage prediction method, device and storage medium | |
| CN116777883B (en) | A method and system for identifying brain tumors in brain magnetic resonance images | |
| CN117197162B (en) | Intracranial hemorrhage CT image segmentation method based on differential convolution | |
| CN117058447B (en) | A pulmonary nodule recognition method and system based on artificial intelligence | |
| CN116823752B (en) | Brain network construction method, system, medium and equipment based on mechanical parameters | |
| CN117911424A (en) | A semi-supervised intracerebral hemorrhage image segmentation method based on double teacher structure | |
| CN114897904B (en) | Method and device for determining fractional flow reserve based on MRA sequence image | |
| CN116563256A (en) | Vascular stenosis rate and embolism grade determining method, device and storage medium | |
| JP7326845B2 (en) | Diffuse case image determination program, system, and method |
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 |