CN113808191A - Automatic quantification and three-dimensional modeling method for focal zone of acute ischemic stroke - Google Patents
Automatic quantification and three-dimensional modeling method for focal zone of acute ischemic stroke Download PDFInfo
- Publication number
- CN113808191A CN113808191A CN202110909875.3A CN202110909875A CN113808191A CN 113808191 A CN113808191 A CN 113808191A CN 202110909875 A CN202110909875 A CN 202110909875A CN 113808191 A CN113808191 A CN 113808191A
- Authority
- CN
- China
- Prior art keywords
- brain
- value
- image
- sequence
- anatomical structure
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 208000032382 Ischaemic stroke Diseases 0.000 title claims abstract description 30
- 238000011002 quantification Methods 0.000 title claims abstract description 16
- 210000004556 brain Anatomy 0.000 claims abstract description 75
- 210000003484 anatomy Anatomy 0.000 claims abstract description 50
- 238000005192 partition Methods 0.000 claims abstract description 47
- 210000003625 skull Anatomy 0.000 claims abstract description 32
- 238000013507 mapping Methods 0.000 claims abstract description 23
- 238000009792 diffusion process Methods 0.000 claims abstract description 14
- 230000003902 lesion Effects 0.000 claims description 38
- 230000036770 blood supply Effects 0.000 claims description 23
- 210000005013 brain tissue Anatomy 0.000 claims description 20
- 230000002490 cerebral effect Effects 0.000 claims description 20
- 238000004364 calculation method Methods 0.000 claims description 16
- 230000004087 circulation Effects 0.000 claims description 11
- 230000000302 ischemic effect Effects 0.000 claims description 9
- 230000002159 abnormal effect Effects 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000003727 cerebral blood flow Effects 0.000 claims description 7
- 210000001638 cerebellum Anatomy 0.000 claims description 6
- 239000010410 layer Substances 0.000 claims description 6
- 239000002356 single layer Substances 0.000 claims description 6
- 230000035515 penetration Effects 0.000 claims description 5
- 230000003993 interaction Effects 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 230000002792 vascular Effects 0.000 claims description 4
- 210000003657 middle cerebral artery Anatomy 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 230000000873 masking effect Effects 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000000638 solvent extraction Methods 0.000 claims description 2
- 210000001652 frontal lobe Anatomy 0.000 claims 3
- 210000003478 temporal lobe Anatomy 0.000 claims 3
- 210000001159 caudate nucleus Anatomy 0.000 claims 2
- 210000000869 occipital lobe Anatomy 0.000 claims 2
- 230000002123 temporal effect Effects 0.000 claims 2
- 210000001103 thalamus Anatomy 0.000 claims 2
- 210000002551 anterior cerebral artery Anatomy 0.000 claims 1
- 210000000133 brain stem Anatomy 0.000 claims 1
- 239000002775 capsule Substances 0.000 claims 1
- 210000001715 carotid artery Anatomy 0.000 claims 1
- 210000004326 gyrus cinguli Anatomy 0.000 claims 1
- 210000001320 hippocampus Anatomy 0.000 claims 1
- 210000001259 mesencephalon Anatomy 0.000 claims 1
- 210000004940 nucleus Anatomy 0.000 claims 1
- 210000001152 parietal lobe Anatomy 0.000 claims 1
- 210000002975 pon Anatomy 0.000 claims 1
- 210000002637 putamen Anatomy 0.000 claims 1
- 206010061216 Infarction Diseases 0.000 description 28
- 230000007574 infarction Effects 0.000 description 25
- 238000002595 magnetic resonance imaging Methods 0.000 description 11
- 210000004204 blood vessel Anatomy 0.000 description 8
- 238000004393 prognosis Methods 0.000 description 8
- 238000011282 treatment Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 238000002560 therapeutic procedure Methods 0.000 description 7
- 238000005481 NMR spectroscopy Methods 0.000 description 6
- 208000006011 Stroke Diseases 0.000 description 6
- 230000010410 reperfusion Effects 0.000 description 6
- 230000002537 thrombolytic effect Effects 0.000 description 6
- 208000026106 cerebrovascular disease Diseases 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 5
- 210000004884 grey matter Anatomy 0.000 description 5
- 239000003086 colorant Substances 0.000 description 4
- 206010058558 Hypoperfusion Diseases 0.000 description 3
- 208000007536 Thrombosis Diseases 0.000 description 3
- 239000003146 anticoagulant agent Substances 0.000 description 3
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 3
- 230000008506 pathogenesis Effects 0.000 description 3
- 210000004885 white matter Anatomy 0.000 description 3
- 206010008190 Cerebrovascular accident Diseases 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 2
- 230000001154 acute effect Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 239000008280 blood Substances 0.000 description 2
- 210000004369 blood Anatomy 0.000 description 2
- 210000001627 cerebral artery Anatomy 0.000 description 2
- 206010008118 cerebral infarction Diseases 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000013156 embolectomy Methods 0.000 description 2
- 238000009472 formulation Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000001990 intravenous administration Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000000750 progressive effect Effects 0.000 description 2
- 208000030090 Acute Disease Diseases 0.000 description 1
- 201000001320 Atherosclerosis Diseases 0.000 description 1
- 206010008089 Cerebral artery occlusion Diseases 0.000 description 1
- 206010069729 Collateral circulation Diseases 0.000 description 1
- 206010021137 Hypovolaemia Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000017531 blood circulation Effects 0.000 description 1
- 201000007293 brain stem infarction Diseases 0.000 description 1
- 230000001269 cardiogenic effect Effects 0.000 description 1
- 210000004004 carotid artery internal Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000010931 gold Substances 0.000 description 1
- 229910052737 gold Inorganic materials 0.000 description 1
- 238000011221 initial treatment Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 208000028867 ischemia Diseases 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 231100000216 vascular lesion Toxicity 0.000 description 1
- 210000002385 vertebral artery Anatomy 0.000 description 1
- 238000013316 zoning Methods 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
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- 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
- G06T2207/10004—Still image; Photographic image
- G06T2207/10012—Stereo images
-
- 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
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Geometry (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Graphics (AREA)
- Software Systems (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses an automatic quantification and three-dimensional modeling method for an acute ischemic stroke focal zone, which comprises the following steps: carrying out skull removing operation on the brain image of the DWI sequence; performing skull removing operation on the brain image of the ADC sequence; creating a new magnetic resonance diffusion weighted image sequence based on the DWI and ADC brain images with the skull removed; dividing a focus area; performing a skull removal operation on the calculated brain image of the T1 sequence; based on the T1 brain image with the skull removed, two brain anatomical regions were made; mapping the identified focus area to two brain anatomical structure partitions, and respectively calculating the focus area volume and the ratio of each brain anatomical structure based on the anatomical structures; and performing three-dimensional modeling based on the mapping result. The invention can three-dimensionally display the distribution form of the focus area in each brain anatomical structure subarea in a short time window and accurately and rapidly calculate the volume and the proportion of the focus area in each brain anatomical structure.
Description
Technical Field
The invention relates to the field of clinical acute ischemic stroke image processing, in particular to a focal zone automatic quantification and three-dimensional modeling method based on different sequences of nuclear magnetic resonance images of brains of patients with acute ischemic stroke.
Background
Acute ischemic stroke is a sudden cerebrovascular disease with a high incidence in the elderly population (Srinivasan, Goyal, Al Azri, & Lum, 2006). Globally, about 74% of patients cannot take care of their lives after the onset of stroke due to loss of limb function or the like, placing a huge economic and labor cost burden on society (Miller et al, 2010).
Reperfusion therapy is currently the internationally accepted treatment most effective for ischemic stroke in the acute phase (Jivan, Ranchod, & Modi, 2013). Reperfusion therapy can be divided into intravenous thrombolytic therapy and mechanical embolectomy. The intravenous thrombolysis treatment is to dissolve thrombus in blood vessel with medicine, mechanically take out thrombus through intravascular intervention treatment, and take out thrombus at the occlusion part with a catheter. Reperfusion therapy recanalizes blood flow, saves brain tissue function, and avoids life-long disability of the patient after stroke onset.
The volume of the focal zone and the distribution of the different areas of cerebral blood supply can greatly influence the decision of the clinician on the reperfusion therapy and the prognosis of the patient. For example, studies have shown that, for example, infarcted lesions are widely distributed in the middle cerebral artery of MCA and the prognosis of mechanical embolectomy is poor (Manceau et al, 2018); infarct zone patterns at different morphological locations in MCA cerebral median arterial infarcts patients were imaged on thrombolytic prognosis in patients (Liu et al, 2015); infarcted areas greater than 70mL had a greater impact on the prognosis of thrombolysis (Tisserand et al, 2016); infarction of the anterior internal carotid artery system and infarction of the posterior basic vertebral artery system have different effects on the prognosis of thrombolysis (Keselman et al, 2020)
Therefore, in practical clinical application scenarios, clinicians usually want to be able to three-dimensionally observe the distribution of the focal region of a patient in each cerebrovascular blood supply region and accurately know the volume and the ratio of the focal region in the cerebrovascular blood supply region, which is instructive for the doctors to accurately make a reperfusion therapy scheme and predict prognosis. Due to the existence of Williams' rings and the difference of actual areas supplied by the same cerebral blood vessel in different individuals, different cerebral blood supply areas have no clear anatomical boundaries, so the accurate volume and the occupation ratio of a focus area in the cerebral blood supply area are difficult to quantify. Past medical research has therefore translated this problem into indirect calculation of the pattern of distribution of focal zones in areas of cerebral blood supply and the volume and ratio of the focal zones in areas of cerebral blood supply by observing their distribution and ratio in areas of the core anatomy that supplies blood to the cerebral blood vessels (Barber, Demchuk, Zhang, & Buchan, 2000).
The abnormally high signal region of the nuclear magnetic resonance DWI sequence is generally considered clinically as the focal region of the stroke episode. At the present stage, the image reading of the acute stage of the patient is completed manually. Because the artificial three-dimensional space imagination and the three-dimensional partition calculation are time-consuming and labor-consuming projects, but the gold treatment time window of the cerebral apoplexy is very narrow, the distribution form of the cerebral apoplexy image report shown by the imaging department in each brain anatomical structure partition of the focal region is rough in description at the present stage, and accurate quantitative calculation is not carried out on the focal region. Therefore, the clinician cannot intuitively and accurately recognize the morphology and the specific volume of the lesion area of the patient in each brain anatomical structure, and the formulation of a treatment scheme and the inaccuracy of prognosis prediction are caused to a certain extent.
Therefore, an algorithm is urgently needed at present, which can accurately and quickly realize skull removal of an image, automatic partition of an image brain anatomical structure, automatic identification of a focus area, mapping calculation of the focus area and different brain anatomical structures and three-dimensional modeling.
Disclosure of Invention
In order to solve the problems of the prior art, the invention provides an automatic quantification and three-dimensional modeling method for an acute ischemic stroke focal region, which is used for displaying the distribution form of the focal region in each brain anatomical structure partition and accurately calculating the volume and the proportion of the focal region in each brain anatomical structure based on the clinical nuclear magnetic resonance image sequence of an acute ischemic stroke patient in a three-dimensional manner, and more accurately and quickly helping clinicians to formulate a reperfusion treatment scheme and predict the prognosis of the patient in an acute disease stage.
The invention adopts the following technical scheme:
an automatic quantification and three-dimensional modeling method for an acute ischemic stroke focal region is characterized in that the treatment initial stage is divided into two Branches (BRANCH) to be synchronously carried out, the calculation results of the two branches are finally gathered to a main step (MASTER100 and MASTER200) to carry out focal region mapping, merging calculation and final three-dimensional modeling, and the two branches are as follows:
BRANCH 100: automatically dividing a brain focal region based on an analog to digital converter (ADC) and a DWI sequence of magnetic resonance imaging (hereinafter, MRI);
BRANCH 200: the MRI T1 sequence brain anatomy is automatically segmented.
Wherein, the BRANCH100 comprises BRANCH101 to BRANCH 104; BRANCH200 includes BRANCH201 and BRANCH 202.
Specifically, the invention relates to an automatic quantification and three-dimensional modeling method for an acute ischemic stroke focal zone, which comprises the following steps:
BRANCH101, which performs skull removing operation on brain images of DWI sequences based on Otsu's method algorithm (Otsu, 1979);
BRANCH102, carrying out skull removing operation on the brain image of the ADC sequence;
a BRANCH103 for creating a new magnetic resonance Diffusion weighted image (DWI ADC Combined Diffusion Sequence, hereinafter referred to as CDsequence) by using the DWI and ADC part image of the brain with the skull removed; the sequence can effectively exclude the influence of the T2 penetration effect;
the BRANCH104 divides the lesion area by using a developed 3D Progression semi-automatic algorithm. Previous deep learning black box algorithms (Zhao et al, 2021) treated a complete three-dimensional focal zone as multiple independent MRI two-dimensional slice focal zones, thus losing much of the three-dimensional spatial information. The 3D progress algorithm treats the focus as a complete three-dimensional individual and combines the characteristics of the sequentially increased Cerebral Blood Flow (CBF) of the ischemic vascular focus area from inside to outside so as to segment the focus area;
BRANCH201, calculating a T1 sequence through DWI and ADC sequences, and carrying out skull removing operation on the brain image of the T1 sequence;
BRANCH202, based on the image of the removed skull from BRANCH201, uses the VoxelMorph algorithm (Balakrishnan et al, 2019) to perform two automatic partitions of the patient's brain anatomy. The first partition is based on the traditional anatomy, and the complete list of anatomies is shown in appendix one, and in a single anatomy, white and gray matter will be distinguished. The second partition is based on the ASPECTS (Alberta Stroke program early CT score) system (Barber et al, 2000), the distribution of the focal region in different anatomical structures indirectly reflects the distribution of the focal region in corresponding blood supply areas of the blood vessels, and the complete anatomical structure list is shown in appendix two;
MASTER100, maps the lesion regions identified by BRANCH104 to two brain anatomical partition maps of BRANCH 202. Respectively calculating the focal region volume and the occupation ratio of each brain anatomical structure according to the anatomical structure lists in the appendix I and the appendix II;
MASTER200, which performs three-dimensional modeling based on the results of a MASTER100 appendix-one mapping, will mark different anatomical regions with different colors, and the focal zone will be superimposed three-dimensionally on the corresponding brain anatomical region.
Preferably, the BRANCH101 specifically includes:
BRANCH101-1, which uses Otsu's method algorithm to calculate a threshold value (threshold) of signal value for distinguishing skull and brain tissue based on the signal value on DWI sequence;
BRANCH101-2, based on the threshold and the original DWI sequence, generating a mask (Mbt) of the brain tissue, specifically, the corresponding value of the area of the original DWI, the signal of which is larger than the threshold, on the Mbt is 1, otherwise, the corresponding value is 0; dividing the Mbt into different three-dimensional connected components by applying a Python masking packet label algorithm, wherein the three-dimensional connected component with the largest number of voxels is a brain tissue mask, and updating the Mbt according to the three-dimensional connected component;
BRANCH101-3, mapping the mask Mbt of the brain tissue to the original DWI sequence to obtain a DWI sequence DWIbt with the skull removed; DWIbt is DWI × Mbt.
Preferably, the BRANCH102 specifically includes:
BRANCH102-1, mapping the mask Mbt of the brain tissue obtained by BRANCH101 to the original ADC sequence to obtain the ADC sequence ADCbt of the removed skull; ADCbt ═ ADC × Mbt.
Preferably, the BRANCH103 specifically includes:
BRANCH103-1, carries on the characteristic scaling (Feature scaling) to DWIbt picture, make all signal values on the picture in the range of [0, 1], the characteristic scaling formulation is:
BRANCH103-2, the Feature scaling (Feature scaling) is carried out on the ADCbt image, all signal values on the image are in the range of [0, 1], and the Feature scaling formula is as follows:
BRANCH103-3, calculating the signal difference image of the DWIbt and ADCbt subjected to feature scaling, and shifting the range section of the signal to ensure that the minimum value of the signal value is a non-negative number:
CDsequence＝DWIbt-ADCbt
CDsequence＝CDsequence-min(CDsequence)
CDsequence＝CDsequence×Mbt。
preferably, the BRANCH104 specifically includes:
BRANCH104-1, automatically identifies the highest point HI of the signal on the lesion slices provided by the physiciani(ii) a Wherein i is a three-dimensional focus number counting variable, and the initial value is 0;
BRANCH104-2, obtaining the extension variable PD adjusted by the doctor on the man-machine interaction interfaceiThe value of (d);
BRANCH104-3, creating focal zoneMask (version Mask i, hereinafter called Lmaki), HIiThe corresponding value of the area on the Lmaki is 1, and the values of the other areas are 0;
BRANCH104-4, a 3D progress algorithm is applied, the algorithm is explained as follows, an extension variable counting variable k is set, and the initial value of k is 0; when k < PDiThen, the following operations are performed:
the extended _ labels algorithm using Python sketch package will be matched with LmaskiMarking the Voxel numerical values corresponding to the three-dimensional connected components with the distances of the voxels (Voxel) with the initial numerical values of 1 as 1 to obtain a mask expand Lesion;
creating a new mask NewLesion-ExpandLesion-Lmaski(ii) a The voxel with the voxel value of 1 on NewLesion is a newly marked three-dimensional connected component voxel in the new cycle of the algorithm;
mapping NewLesion to CDsequence; ExpandThreshold ═ max (CDsequence × newversion); ExpandThreshold is the maximum signal value of the newly labeled voxel on the created new magnetic resonance diffusion-weighted image CDsequence;
relabeling voxels with a CDsequence value less than Expandthreshold as 0 on the mask NewLesion; the region with the value of 1 in NewLesion is the voxel to be marked as a new focal zone in the new cycle;
updating focus area mask Lmaski＝Lmaski+NewLesion；
Counting the value of a variable k and adding 1;
BRANCH104-5, doctor increases PDiCounting the values, comparing the recognition result with the DWI original image, and automatically repeating the algorithm in BRANCH104-4 to continuously update LmaskiIdentifying that the focal zone is continuously extending outward; when the abnormal area of the signal covered by the focus area is identified, the doctor stops adjusting and increasing PDiNumerical value, the identification of a single three-dimensional focal zone is finished;
BRANCH104-6, when PD is increasediThe numerical value enables the doctor to reduce the PD when the identified focus area covers the non-signal abnormal area on the original DWI imageiNumerical values were such that they covered only the signal abnormality region, and "next independent lesion identification" was clicked; counting the value of a variable i and adding 1; automaticUpdating and identifying signal peak HI on residual lesion slicei；
BRANCH104-7, automatically repeat BRANCH104-2 to BRANCH 104-6;
BRANCH104-8, when the identification area covers all the signal abnormal areas on the original DWI, the three-dimensional focus area identification is finished, and finally the focus area mask Lmak ∑ Lmaki。
Preferably, the BRANCH201 specifically includes:
BRANCH201-1, calculating a T2 sequence (a DWI sequence with a b value of 0) according to the DWI and the ADC sequence, mapping the mask Mbt of the brain tissue obtained by BRANCH101 to a T2 sequence to obtain a T2 sequence T2bt with the skull removed; t2bt ═ T2 × Mbt; the T2bt image is feature scaled so that all signal values on the image are within the 0, 1 range. Carrying out inverse conversion on the T2bt picture signal value to obtain a T1 sequence T1bt with skull removed; t1 bt-1-T2 bt.
Preferably, the BRANCH202 specifically includes:
BRANCH202-1, obtaining a Template (Brain T1 Template, hereinafter referred to as Template) of an adult Brain T1 image from an open source platform GitHub of VoxelMorph; the Template is automatically synthesized by a computer based on the characteristics of 7829 adult brain T1 images;
BRANCH202-2, partition the Template according to the subdivision in appendix one by using BrainSuite software to generate partition mask Template 1; in template 1, voxels corresponding to the same partition in appendix one are all assigned the same positive integer; voxels that do not belong to a partition in appendix one are assigned a value of 0;
BRANCH202-3, partition the Template according to the partition in appendix two by using BrainSuite software to generate partition mask Template 2; in template 2, voxels corresponding to the same partition in appendix two are all assigned the same positive integer; voxels that do not belong to the partition in appendix two are assigned a value of 0;
BRANCH202-4, preprocessing the T1bt image of the patient to make the voxel signal value interval and the size of the image three-dimensional matrix the same as the Template: the size of the T1bt three-dimensional matrix is 256 × 256 × 256 by using a mathematical Interpolation (Interpolation) algorithm, and the matrix is cut to 160 × 192 × 224. Finally, we perform Feature scaling (Feature scaling) on the T1bt image to make all signal values in the image within the range of [0, 1], and the Feature scaling formula is:
BRANCH202-5, using a VoxelMorph image registration network to align Template and T1 bt. In the registration process, templates 1 and 2 are registered to T1bt to form T1btSeg1 and T1btSeg2, respectively. T1btSeg1 is the T1bt partition mask generated according to annex one, and T1btSeg2 is the T1bt partition mask generated according to annex two.
Preferably, the MASTER100 specifically includes:
MASTER100-1 applies mathematical Interpolation (Interpolation) algorithm to make Lmak three-dimensional matrix size 256 × 256 and cut down to 160 × 192 × 224.
MASTER100-2, calculating the volume represented by a single voxel after applying a mathematical interpolation algorithm
Wherein, spacing Betwensballs represents the longitudinal distance between the center of one layer of Slice of the original DWI image and the center of the adjacent layer; PixelSpacing represents the distance of the pixels of a single layer of an original DWI image; NbSlice represents the slice level of the original DWI image;
256 in (d) represents the new number of layers after interpolation; NbPixel represents the number of single-side pixels of a single-layer slice of the original DWI image;
256 in the series is the number of unilateral pixels of the single-layer slice after interpolation calculation;
MASTER100-3, calculate the total lesion volume VLesion:
VLesion＝Vvoxel×NbVoxelLesion
wherein NbVoxelLesion represents the number of voxels with a value of 1 on the Lmask after interpolation calculation;
MASTER100-4, mapping Lmak to T1btSeg1, as follows:
T1btSeg1Lesion＝Lmask×T1btSeg1
the MASTER100-5 respectively calculates the focal region volume and the occupation ratio of each brain anatomical structure according to the anatomical structure list in the appendix I; assuming an anatomical structure j, whose corresponding voxel has a value j in T1btSeg1 version, the volume of the focal zone in the anatomical structure j is:
VLesionj＝Vvoxel×NbVoxelLesionj
NbVoxelLesion j represents the number of voxels with a voxel value of j in T1btSeg1 Lesion;
the proportion of the focal zone in the anatomical structure j is:
wherein NbVoxelj represents the number of voxels with a voxel value of j in T1btSeg 1;
MASTER100-6, maps Lmask to T1btSeg2 as follows:
T1btSeg2Lesion＝Lmask×T1btSeg2
MASTER100-7, respectively calculating the focal zone volume and the ratio of each brain anatomical structure according to the anatomical structure list in appendix II; assuming an anatomical structure k, whose corresponding voxel has a value k in the T1btSeg2 version, the volume of the focal zone in the anatomical structure k is:
VLesionk＝Vvoxel×NbVoxelLesionk
wherein NbVoxelLesion k represents the number of voxels with a voxel value of k in T1btSeg2 Lesion;
the proportion of the focal zone in the anatomical structure k is:
where NbVoxelk represents the number of voxels with a voxel value of k in T1btSeg 2.
Preferably, the MASTER200 specifically includes:
MASTER200-1, applies mathematical interpolation algorithm to make the size of Mbt three-dimensional matrix 256 × 256 and cut down to 160 × 192 × 224. Based on a mask Mbt of a brain tissue, a legosurface module in a source code of open source software vedo is used for outlining the brain of a patient, voxels with the same voxel value in the Mbt are displayed in a unified color in 3D modeling, and a 3D model BrainVolume is obtained;
MASTER200-2, based on the mapping result T1btSeg1 version of MASTER100-4, using a legosurface module in a source code of open source software vedo to outline an anatomical region involved in a focus area, wherein voxels with the same voxel value in the T1 btSeglversion are displayed in uniform color in 3D modeling, and obtaining a 3D model SegVolume;
the MASTER200-3 is used for reducing the transparency of the 3D model Brainvolume by using a shrink module in the source code of vedo, and overlapping and displaying the 3D model Segvolume on the 3D model Brainvolume;
the MASTER200-4 is based on a focus area mask Lmask of BRANCH104-8, a focus area is outlined by utilizing a legosurface module in a source code of open source software vedo, voxels with the same voxel value in the Lmask are displayed in a unified color in 3D modeling, and a 3D model Lesionvolume is obtained;
and the MASTER200-5 is used for reducing the transparency of the 3D model Segvolume by using a shrink module in the source code of vedo, and overlapping and displaying the 3D model Lesionvolume on the 3D model Segvolume.
The invention has the following beneficial effects:
(1) according to the invention, the interpretability of the artificial intelligence model is greatly improved through characteristic engineering closely combined with clinical knowledge of neurology, the acceptance of clinicians is improved, the working efficiency of the clinicians is better improved, and the working intensity of the clinicians is reduced;
(2) the invention can three-dimensionally display the distribution form of the lesion infarction core area in each brain anatomical structure partition based on the clinical nuclear magnetic resonance image sequence of the acute ischemic stroke patient and accurately and rapidly calculate the volume and the proportion of the infarction area in each brain anatomical structure.
The present invention will be described in further detail with reference to the accompanying drawings and embodiments, but the method for automatically quantifying and three-dimensional modeling of an acute ischemic stroke focal zone according to the present invention is not limited to the embodiments.
Drawings
FIG. 1 is a flow chart of a method for automatic quantification and three-dimensional modeling of an acute ischemic stroke focal zone according to an embodiment of the present invention;
FIG. 2 is an anatomical structure list diagram of appendix one;
FIG. 3 is an anatomical structure list diagram of appendix two;
FIG. 4 is a schematic diagram of a section of a patient RJ with a lesion being circled on a human-computer interface;
FIG. 5 shows the signal peaks of RJ lesion areas of a patient identified by the algorithm and marked by circles;
FIG. 6 is a schematic diagram illustrating the progressive diffusion of the identified focal zone of patient RJ from the high signal zone to the low signal zone as the extension variable PD is progressively increased by the doctor;
FIG. 7 is a schematic diagram of the final identified focal zone of a patient RJ;
FIG. 8 is a schematic illustration of a section of a multi-infarct patient WS containing a lesion circled on a human-machine interface;
FIG. 9 shows the signal peaks in the lesion area of the multi-infarcted patient WS identified by the algorithm and marked with a circle;
FIG. 10 is a partial schematic view of the progressive diffusion of the identified focal zone of the patient WS from the high signal region to the low signal region as the extension variable PD is progressively increased by the physician;
fig. 11 is a view showing that the PD value is enlarged to cover the normal brain tissue in the identification region, and the doctor presses "identification of the next independent lesion", and identification of a single lesion a is completed;
FIG. 12 is a diagram illustrating automatic identification of the remaining signal peaks, and is circled;
FIG. 13 is a schematic view of the newly identified focal zone B gradually diffusing from the high signal zone to the low signal zone as the extension variable PD is gradually increased by the physician;
FIG. 14 is a schematic diagram of the ultimate recognition of focal zone (A + B) by the multi-infarct patient WS;
fig. 15 is a three-dimensional lesion volume map generated based on a raw MRI two-dimensional image of patient WS that can be viewed from multiple angles.
Detailed Description
The invention is further described below by means of specific embodiments. It should be noted that the specific embodiments described herein are only for convenience of describing and explaining specific embodiments of the present invention, and are not intended to limit the present invention.
Referring to fig. 1, in the automatic quantification and three-dimensional modeling method for the focal zone of acute ischemic stroke, the initial treatment stage is divided into two Branches (BRANCH) for synchronous operation, and the calculation results of the two algorithm branches are finally summarized to a main step (MASTER) for focal zone mapping, merging calculation and final three-dimensional modeling.
BRANCH 100: automatic division of brain focal zone based on magnetic resonance imaging (hereinafter referred to as MRI) ADC and DWI sequence
BRANCH 200: the MRI T1 sequence brain anatomy is automatically segmented. d1) BRANCH100 is subdivided into the following steps:
BRANCH 101: the brain images of the DWI sequence were subjected to a skull removal operation based on Otsu's method algorithm (Otsu, 1979).
BRANCH 102: mapping the brain tissue mask (mask) obtained by BRANCH101 for removing the skull onto the ADC sequence, and carrying out skull removing operation on the brain image of the ADC sequence.
BRANCH 103: a new magnetic resonance Diffusion weighted image (hereinafter referred to as DWI ADC Combined Diffusion Sequence) is created using the DWI and ADC brain images with the skull removed. The sequence can effectively eliminate the influence of the penetration effect of T2.
BRANCH 104: and dividing the lesion area by utilizing a developed 3D Progression semi-automatic algorithm. Previous deep learning black box algorithms (Zhao et al, 2021) treated a complete three-dimensional focal zone as multiple independent MRI two-dimensional slice focal zones, thus losing much of the three-dimensional spatial information. The 3D progress algorithm designed by the invention treats the focus as a complete three-dimensional individual and combines the characteristic that the Cerebral Blood Flow (CBF) of the ischemic vascular focus area is sequentially increased from inside to outside so as to segment the focus area.
(2) BRANCH200 is subdivided into the following steps:
BRANCH 201: calculating a T1 sequence through DWI and ADC sequences, mapping a skull-removed brain tissue mask (mask) obtained by BRANCH101 onto a T1 sequence, and carrying out skull removing operation on a brain image of a T1 sequence
BRANCH 202: based on the image of the BRANCH201 removed skull, the voxel morphh algorithm (Balakrishnan, Zhao, Sabuncu, Guttag, & Dalca,2019) was used to perform two automatic partitions of the patient's brain anatomy. The first partition is based on the traditional anatomy, and the complete anatomical structure list is shown in appendix one of fig. 2, where white and gray matter will be distinguished in a single anatomical structure. The second partition is based on the aspects (alberta stroke program early CT score) scoring system (Barber et al, 2000), and the distribution of the focal zone in different anatomical structures indirectly reflects the distribution in the corresponding blood-supply region of the vessel, and the complete anatomical structure list is shown in appendix two of fig. 3.
(3) BRANCH calculation results of BRANCH100 and BRANCH200 are summarized to a trunk MASTER, and the trunk steps can be subdivided into the following steps:
MASTER 100: the lesion regions identified by BRANCH104 are mapped to two brain anatomy partition maps of BRANCH 202. The focal zone volume and the proportion of each brain anatomical structure are respectively calculated according to the anatomical structure lists in the appendix one and the appendix two
MASTER 200: three-dimensional modeling is performed based on the results of a MASTER100 appendix-mapping, different anatomical zones will be marked with different colors, and the focal zone will be superimposed three-dimensionally on the corresponding brain anatomical zone.
On one hand, the automatic quantification and three-dimensional modeling method for the acute ischemic stroke focal zone greatly improves the interpretability of an artificial intelligence model through characteristic engineering closely combined with clinical knowledge of neurology, improves the acceptance of clinicians, better improves the working efficiency of the clinicians and reduces the working intensity of the clinicians.
In particular, traditional medical big data methods are based entirely on black box models, resulting in clinician distrust and non-adoption of the methods. The method of the invention is closely related to imaging, stroke pathogenesis, vascular anatomy and brain anatomy, and specifically comprises the following steps:
first, in step BRANCH103, a new magnetic resonance Diffusion weighted image (DWI ADC Combined Diffusion Sequence) is created using the DWI and ADC sequences to exclude the effects of the T2 penetration effect. Due to the penetration effect of T2, the highlights on the DWI sequence are not necessarily diffusion limited regions (focal zones) but are likely to be high signals originating from T2. To further determine whether the DWI highlight is a focal zone, it is necessary to verify whether the ADC sequence values corresponding to the DWI highlight are in the low signal range. On magnetic resonance diffusion-weighted images created based on DWI and ADC sequences, only regions with both high signal values on the DWI and low signal values on the ADC will show a diffusion-limited high signal.
Secondly, in step BRANCH104, a self-developed semi-automatic 3D progress algorithm is applied to segment the infarct focus area. The algorithm is better combined with the pathogenesis of acute ischemic stroke. Acute ischemic stroke is caused by occlusion of a cerebral artery. Cerebral artery occlusion causes ischemia in the blood supply area, and cerebral infarction occurs after a certain time limit. Due to the collateral circulation, the ischemic area of blood supply caused by occlusion of the cerebral arteries is still supplied with a partial supply of blood, thus leading to incomplete infarct focus: the Infarct focus consists of the central core Infarct zone (Infarct core) and ischemic Penumbra (Penumbra). There are also areas of Benign hypoperfusion around the penumbra (Benign oligemia). The ischemic vascular lesion area is sequentially elevated in Cerebral Blood Flow (CBF) from inside to outside, from the core infarct area to the benign hypoperfusion area (Bandera et al, 2006). It can be concluded that the signal strength in the focal zone should decrease from inside to outside in the DWI ADC Combined Diffusion Sequence. Furthermore, from the point of view of mathematical graph theory, each ischemic infarct focal zone is a three-dimensional Connected Component (Connected Component) extending along the direction of the culprit vessel and its branches. By combining the geometric characteristics of the ischemic infarction focal zone and the image signal characteristics deduced by pathogenesis, the algorithm takes the focus signal highest point on the DWI ADC Combined Diffusion Sequence as the center of the three-dimensional connected component, and gradually extends the three-dimensional connected component to the outer low signal zone until the three-dimensional connected component completely covers the signal abnormal zone.
The working logic of the algorithm is detailed as follows by taking the nuclear magnetic resonance image of the brain of an RJ patient with one post-cycle infarction as an example: in practical clinical applications, the physician only needs to place the section containing the focal zone in the circle shown in FIG. 4 on the human-computer interaction computer interface provided by the present invention. The algorithm of the present invention will automatically identify the highest point of the signal in the three-dimensional spatial space of the focal zone. The focal signal maxima for the patient RJ appear in the circled area of the second slice as shown in figure 5.
Then, the doctor can adjust the single numerical extension variable Progression Degree (PD) on the human-computer interaction interface to enable the focus three-dimensional connected component to gradually extend from the highest signal to the outer low signal area. Referring to fig. 6, as the PD values increased, the focal zone was identified as gradually extending from the central core infarct zone to the ischemic penumbra and the benign hypoperfusion zone. Referring to fig. 7, when the three-dimensional connected component completely covers the signal abnormality region, the doctor may stop adjusting the PD variable and the lesion recognition is completed.
If the patient has only a single three-dimensional connected component stereo focus (common infarction caused by atherosclerosis), doctors can make the identification area cover all focus areas by continuously increasing the PD value. If a patient has a plurality of independent three-dimensional connected component stereo lesions (common in cardiogenic infarction), the identification area can cover normal brain tissues between different stereo lesion areas by simply increasing the PD value. Therefore, when the PD value is increased to cover the identified region with normal brain tissue, the clinician simply clicks "next independent lesion identification" on the human-computer interface, the system will automatically identify the signal peak in the circle in the remaining red frame, and repeat the above identification, and fig. 8 to 14 illustrate the process in detail by taking a multi-infarct patient WS as an example.
Finally, in BRANCH202, the invention simulates the ASPECTS score to quantitatively partition the blood supply area of the brain, and the distribution of the focus area in different anatomical structures indirectly reflects the distribution of the focus area in the corresponding blood supply area. The hitherto more common quantification of Brain partitions (Brain Atlas) is generally based on traditional anatomical or functional partitioning of the Brain, such as the presently disclosed Brain partitions Data set 2012MICCAI Multi-Atlas laboratory change Data or Hammers adult atriases. The invention designs the brain blood vessel blood supply area quantitative partition according to the ASPECTS score according to the characteristics of the cerebrovascular disease. The blood vessel blood supply area quantitative partition map can contain the relevant morbidity information of responsible blood vessels of patients, and accords with the logic of clinical diagnosis and treatment of cerebrovascular diseases.
On the other hand, the automatic quantification and three-dimensional modeling method for the focal zone of the acute ischemic stroke can three-dimensionally display the distribution form of the focal infarction core zone in each brain anatomical structure partition based on the clinical nuclear magnetic resonance image sequence of the acute ischemic stroke patient and accurately and rapidly calculate the volume and the proportion of the infarct zone in each brain anatomical structure.
The following experiment was performed based on the WS case of acute ischemic stroke patient in FIG. 8.
By utilizing the method provided by the invention, a three-dimensional lesion zoning image (shown in figure 15) which can be watched in multiple angles is generated based on the original MRI two-dimensional image of a patient, different brain anatomical regions in a 3D image are marked by different colors, and an irregular monoblock-shaped geometric body marked with colors is a core lesion zone which is automatically identified by an algorithm and presented in 3D. The method of the invention also calculates the following quantization indexes: the total infarct volume of the patient was 4.87 cubic centimeters.
According to the partition of the traditional anatomical structure in appendix I, the volume of the infarction of the grey matter of the cerebellum at one side is 2.63 cubic centimeters and accounts for 4.79 percent of the total volume of the grey matter of the cerebellum at one side; the volume of white matter infarction of the cerebellum at one side is 0.64 cubic centimeter, which accounts for 3.50 percent of the total volume of the grey matter of the cerebellum at one side; the volume of the cerebral infarction is 1.60 cubic centimeters and accounts for 6.10 percent of the total volume of the cerebellum.
According to the area division of the blood vessel supply area in the appendix II, the infarct volume of one cerebellar hemisphere is 3.27 cubic centimeters and accounts for 4.48 percent of the total volume of one cerebellar hemisphere; the volume of the pontine infarction is 1.60 cubic centimeters and accounts for 6.10 percent of the total volume of the pontine.
The above method runs on a computer (configured as Processor 2.6GHz, Intel Core i5, memory 8Go 1600MHz DDR3) for 150 seconds.
At the same time, the clinician (three years of neurology clinical experience) re-evaluated the image of this patient using 3D Slicer software. The artificially delineated focal region and the computer automatically delineated focal region have an average Dice Score of 0.910 at different anatomical/vessel regions. The time taken to manually delineate the focal zone and calculate the above quantitative index was 24 minutes (1440 seconds).
To further quantify the performance of the invention, 185 acute ischemic stroke patients were taken MRI images from a hospital before thrombolytic therapy. Sensitive personal information for all patients has been anonymized by the hospital side and the data collection process has been approved by the hospital ethics committee. Of 185 patients, 68% were infarcted in the area of cerebral anterior circulation, 28% were infarcted in the area of cerebral posterior circulation, and 4% were infarcted in watershed PCA-MCA.
The invention aims to three-dimensionally display the distribution form of a focus area in each brain anatomical structure region in a time window which is as short as possible and accurately and quickly calculate the volume and the proportion of the focus area in each brain anatomical structure. The index for evaluating the model should therefore consist of the model running time and the calculation accuracy of the volume and proportion of the lesion area in the respective brain anatomy. The invention designs the following evaluation indexes:
suppose the brain head image of M (185) patients is segmented into N (29, 12 in appendix one and 17 in appendix two) brain anatomical partitions (the partitions are described in the BRANCH202 step), P is the patient set, SiSet of patient anatomy partitions for i:
P＝{pi}i∈[0，M-1]
Si＝{sij}j∈[0，N-1]
lija value of 1 if the lesion area of patient i is manually judged to be present in anatomy j, otherwise a value of 0; segmijA set of pixels representing a focal zone in all patient i anatomy j manually marked using conventional methods; segaijA set of pixels representing the focal zone in all patient i anatomies j marked using the present invention; tmiThree-dimensional modeling and accurate quantification of the time (in seconds) of partition calculation are carried out on a focus image of a patient by a clinician through 3D Slicer software; ta isiThe time (taking seconds as a unit) for three-dimensional modeling and accurate quantitative partition calculation of the focus image of the patient is calculated by applying the method;
the results calculated by manually applying the traditional method are all regarded as the gold standard, and are defined as follows:
for each patient image processing assessment, define:
DCi denotes the average degree of overlap of the anatomical structures of each brain between the manually marked infarct zone and the lesion zone marked using an algorithm for patient i. tpentaityiIs a time penalty item; performancei∈[-1，1]. Performance under Ideal conditionsiA maximum value of 1 is reached, at which point DCitPenalty of 1iIs 0, i.e. represents the time in which the infinity approaches 0 seconds, the model meter of the inventionThe same lesion partition quantification result as the gold standard was calculated. On the contrary, if PerformanceiThe more close to-1, the result of the comprehensive evaluation index of the model is not ideal.
The method is applied to a data set for evaluation, the experimental result is shown in table 1, and the experimental data format is mean +/-SD.
TABLE 1
According to the experimental result, the comprehensive evaluation index of the algorithm is optimal when the focus area is positioned in the posterior cycle. Compared with a posterior circulation blood supply area, the anatomical structure of the anterior circulation blood supply area is complex, so that the DC value of the algorithm is lower than that of the posterior circulation, and the overall evaluation index of the algorithm is slightly worse than that of the posterior circulation.
It will be understood that modifications and variations can be resorted to by those skilled in the art, and that all such modifications and variations are intended to be included within the scope of the invention as defined by the appended claims.
Claims (9)
1. An automatic quantification and three-dimensional modeling method for an acute ischemic stroke focal zone is characterized by comprising the following steps:
BRANCH101, which performs skull removing operation on the brain image of the DWI sequence;
BRANCH102, carrying out skull removing operation on the brain image of the ADC sequence;
BRANCH103, based on DWI and ADC brain image of removed skull, creating new magnetic resonance diffusion weighted image sequence CDsequence to exclude the effect of T2 penetration effect;
BRANCH104, regarding the focus as a complete three-dimensional individual, and combining the characteristics that the cerebral blood flow CBF of the ischemic vascular focal zone is sequentially increased from inside to outside to segment the focal zone;
BRANCH201, calculating a T1 sequence through DWI and ADC sequences, and carrying out skull removing operation on the brain image of the T1 sequence;
BRANCH202, based on T1 brain image of the removed skull, performs two kinds of brain anatomy partitioning;
the MASTER100 is used for mapping the lesion area identified by BRANCH104 to two brain anatomical structure partitions of BRANCH202, and respectively calculating the lesion area volume and the occupation ratio of each brain anatomical structure based on the anatomical structure list;
the MASTER200 carries out three-dimensional modeling based on the mapping result of the MASTER100, different anatomical structure partitions use different labels, and a focus area is stereoscopically superposed on the corresponding brain anatomical structure partition.
2. The method for automatically quantifying and three-dimensional modeling of an acute ischemic stroke focal zone as claimed in claim 1, wherein the BRANCH101 specifically comprises:
BRANCH101-1, which uses Otsu's method algorithm to calculate a threshold value threshold of signal value capable of distinguishing skull and brain tissue based on the signal value on DWI sequence;
BRANCH101-2, generating a mask Mbt of the brain tissue based on the threshold and the original DWI sequence, namely, the corresponding value of the area, in which the signal of the original DWI sequence is larger than the threshold, on the Mbt is 1, otherwise, the corresponding value is 0; dividing the Mbt into different three-dimensional connected components by applying a Python masking packet label algorithm, wherein the three-dimensional connected component with the largest number of voxels is a brain tissue mask, and updating the Mbt according to the three-dimensional connected component;
BRANCH101-3, mapping the mask Mbt of the brain tissue to the original DWI sequence to obtain a DWI sequence DWIbt with the skull removed; DWIbt is DWI × Mbt.
3. The method for automatically quantifying and three-dimensional modeling of an acute ischemic stroke focal zone as claimed in claim 2, wherein the BRANCH102 specifically comprises:
mapping the mask Mbt of the brain tissue obtained by BRANCH101 to the original ADC sequence to obtain the ADC sequence ADCbt with the skull removed; ADCbt ═ ADC × Mbt.
4. The method for automatically quantifying and three-dimensional modeling of an acute ischemic stroke focal zone as claimed in claim 3, wherein the BRANCH103 specifically comprises:
BRANCH103-1, carry on the characteristic scaling to DWIbt picture, make all signal values on the picture in the range of [0, 1 ];
BRANCH103-2, which performs characteristic scaling on the ADCbt image to make all signal values on the image within the range of [0, 1 ];
and the BRANCH103-3 calculates a signal difference image of the DWIbt and the ADCbt subjected to feature scaling, translates the range interval of the signal values to ensure that the minimum value of the signal values is a non-negative number, and obtains a new magnetic resonance diffusion weighted image sequence CDsequence.
5. The method for automatically quantifying and three-dimensional modeling of an acute ischemic stroke focal zone as claimed in claim 4, wherein the BRANCH104 specifically comprises:
BRANCH104-1, automatically identifies the highest point HI of the signal on the lesion slices provided by the physiciani(ii) a Wherein i is a three-dimensional focus number counting variable, and the initial value is 0;
BRANCH104-2, obtaining the value of the extension variable PDi adjusted by the doctor on the man-machine interaction interface;
BRANCH104-3, creating focal zone mask Lmaki，HIiIn the region of LmakiThe corresponding numerical value is 1, and the numerical values of the rest areas are 0;
BRANCH104-4, setting an initial value of an extension variable counting variable k, wherein k is 0; when k < PDiThen, the following operations are performed:
the extended _ labels algorithm using Python sketch package will be matched with LmaskiMarking the Voxel numerical values corresponding to the three-dimensional connected components with the Voxel distance of 1 and the initial numerical value of l as 1 to obtain a mask expand Lesion;
creating a new mask NewLesion-ExpandLesion-Lmaski(ii) a The voxel with the voxel value of 1 on NewLesion is a newly marked three-dimensional connected component voxel in the new cycle of the algorithm;
mapping NewLesion to CDsequence; ExpandThreshold ═ max (CDsequence × newversion); ExpandThreshold is the maximum signal value of the newly labeled voxel on the created new magnetic resonance diffusion-weighted image CDsequence;
relabeling voxels with a CDsequence value less than Expandthreshold as 0 on the mask NewLesion; the region with the value of 1 in NewLesion is the voxel to be marked as a new focal zone in the new cycle;
updating focus area mask Lmaski＝Lmaski+NewLesion；
Counting the value of a variable k and adding 1;
BRANCH104-5, increasing PDiCounting the values, comparing the recognition result with the DWI original image, and automatically repeating the algorithm in BRANCH104-4 to continuously update LmaskiIdentifying that the focal zone is continuously extending outward; stopping adjusting and increasing PD when recognizing that the focus area completely covers the abnormal area of the signaliNumerical value, the identification of a single three-dimensional focal zone is finished;
BRANCH104-6, when PD is increasediThe numerical value makes the PD be reduced when the identified focus area covers the non-signal abnormal area on the original DWI imageiThe numerical value only covers the signal abnormal area, and a 'next independent focus identification' instruction is obtained; counting the value of a variable i and adding 1; automatic updating HI for identifying highest point of signal on residual focus sectioni；
BRANCH104-7, automatically repeat BRANCH104-2 to BRANCH 104-6;
BRANCH104-8, when the identification area covers all the signal abnormal areas on the original DWI, the three-dimensional focus area identification is finished, and finally the focus area mask Lmak ∑ Lmaki。
6. The method for automatically quantifying and three-dimensional modeling of an acute ischemic stroke focal zone as claimed in claim 5, wherein the BRANCH201 specifically comprises:
calculating a T2 sequence according to the DWI and the ADC sequence, mapping the mask Mbt of the brain tissue obtained by BRANCH101 to a T2 sequence to obtain a T2 sequence T2bt with the skull removed, wherein T2bt is T2 multiplied by Mbt; scaling the features of the T2bt image to bring all signal values on the image within the [0, 1] range; carrying out inverse conversion on the T2bt picture signal value to obtain a T1 sequence T1bt with skull removed, wherein T1bt is 1-T2 bt; here, the T2 sequence represents a DWI sequence with a b value of 0.
7. The method for automatically quantifying and three-dimensional modeling of an acute ischemic stroke focal zone as claimed in claim 6, wherein the BRANCH202 specifically comprises:
BRANCH202-1, obtaining Template of T1 image of adult brain;
BRANCH202-2, divide Template according to the first anatomical structure and produce the partition mask Template 1; in template 1, the voxels corresponding to the same partition are all assigned with the same positive integer, otherwise, the voxels are assigned with 0; wherein the first anatomical structure comprises frontal lobe, parietal lobe, temporal lobe, occipital lobe, islet lobe, cingulate gyrus, thalamus, caudate nucleus, pisiform nucleus, hippocampus, brainstem, and cerebellum;
BRANCH202-3, divide Template according to the second anatomical structure and produce the Template of the partition mask 2; in template 2, the voxels corresponding to the same partition are all assigned with the same positive integer, otherwise, the voxels are assigned with 0; wherein the second anatomical structure consists of a blood supply area of a middle cerebral artery in the anterior cerebral circulation, a blood supply area of an anterior cerebral artery in the anterior cerebral circulation, a blood supply area of a anterior carotid artery and a blood supply area of a posterior cerebral circulation; the cerebral anterior circulation middle cerebral artery blood supply area consists of a caudate nucleus, a putamen, an inner capsule, an island leaf, a frontal lobe island cover, an anterior temporal leaf, a posterior temporal leaf, an upper frontal lobe island cover, an upper anterior temporal lobe and an upper posterior temporal lobe; the cerebral posterior circulation blood supply area consists of thalamus, occipital lobe, midbrain, pons and cerebellar hemisphere;
BRANCH202-4, preprocessing the T1bt image of the patient to make the voxel signal value interval and the size of the image three-dimensional matrix the same as the Template;
BRANCH202-5, using a VoxelMorph image registration network to align Template and T1 bt; in the registration process, templates 1 and 2 are registered to T1bt to form T1btSeg1 and T1btSeg2, respectively; t1btSeg1 is the T1bt partition mask generated from the first anatomical list, and T1btSeg2 is the T1bt partition mask generated from the second anatomical list.
8. The method according to claim 7, wherein the MASTER100 specifically comprises:
the MASTER100-1 adopts a mathematical interpolation algorithm to make the size of the Lmak three-dimensional matrix be 256 multiplied by 256 and cut down to 160 multiplied by 192 multiplied by 224;
MASTER100-2, calculating the volume represented by a single voxel after applying a mathematical interpolation algorithm
Wherein, spacing Betwensballs represents the longitudinal distance between the center of one layer of Slice of the original DWI image and the center of the adjacent layer; PixelSpacing represents the distance of the pixels of a single layer of an original DWI image; NbSlice represents the slice level of the original DWI image;
256 in (d) represents the new number of layers after interpolation; NbPixel represents the number of single-side pixels of a single-layer slice of the original DWI image;
256 in the series is the number of unilateral pixels of the single-layer slice after interpolation calculation;
MASTER100-3, calculate the total lesion volume VLesion:
VLesion＝Vvoxel×NbVoxelLesion
wherein NbVoxelLesion represents the number of voxels with a value of 1 on the Lmask after interpolation calculation;
MASTER100-4, mapping Lmak to T1btSeg1, as follows:
T1btSeg1Lesion＝Lmask×T1btSeg1
the MASTER100-5 is used for respectively calculating the focal zone volume and the occupation ratio of each brain anatomical structure according to the first anatomical structure; assuming an anatomical structure j, whose corresponding voxel has a value j in T1btSeg1 version, the volume of the focal zone in the anatomical structure j is:
VLesionj＝Vvoxel×NbVoxelLesionj
NbVoxelLesion j represents the number of voxels with a voxel value of j in T1btSeg1 Lesion;
the proportion of the focal zone in the anatomical structure j is:
wherein NbVoxelj represents the number of voxels with a voxel value of j in T1btSeg 1;
MASTER100-6, maps Lmask to T1btSeg2 as follows:
T1btSeg2Lesion＝Lmask×T1btSeg2
MASTER100-7, respectively calculating the focal zone volume and the ratio of each brain anatomical structure according to the second anatomical structure; assuming an anatomical structure k, whose corresponding voxel has a value k in the T1btSeg2 version, the volume of the focal zone in the anatomical structure k is:
VLesionk＝Vvoxel×NbVoxelLesionk
wherein NbVoxelLesion k represents the number of voxels with a voxel value of k in T1btSeg2 Lesion;
the proportion of the focal zone in the anatomical structure k is:
where NbVoxelk represents the number of voxels with a voxel value of k in T1btSeg 2.
9. The method for automatic quantification and three-dimensional modeling of an acute ischemic stroke focal zone as claimed in claim 8, wherein the MASTER200 specifically comprises:
the MASTER200-1, which uses a mathematical interpolation algorithm to make the size of the Mbt three-dimensional matrix be 256 multiplied by 256 and cut down to 160 multiplied by 192 multiplied by 224; based on a mask Mbt of a brain tissue, a legosurface module in a source code of open source software vedo is used for outlining the brain of a patient, voxels with the same voxel value in the Mbt are displayed in a unified color in 3D modeling, and a 3D model BrainVolume is obtained;
MASTER200-2, based on the mapping result T1btSeg1 version of MASTER100-4, using a legosurface module in a source code of open source software vedo to outline an anatomical region involved in a focus area, wherein voxels with the same voxel value in the T1bt1 version are displayed in uniform color in 3D modeling, and obtaining a 3D model SegVolume;
the MASTER200-3 is used for reducing the transparency of the 3D model Brainvolume by using a shrink module in the source code of vedo, and overlapping and displaying the 3D model Segvolume on the 3D model Brainvolume;
the MASTER200-4 is based on a focus area mask Lmask of BRANCH104-8, a focus area is outlined by utilizing a legosurface module in a source code of open source software vedo, voxels with the same voxel value in the Lmask are displayed in a unified color in 3D modeling, and a 3D model Lesionvolume is obtained;
and the MASTER200-5 is used for reducing the transparency of the 3D model Segvolume by using a shrink module in the source code of vedo, and overlapping and displaying the 3D model Lesionvolume on the 3D model Segvolume.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110909875.3A CN113808191B (en) | 2021-08-09 | 2021-08-09 | Automatic quantification and three-dimensional modeling method for focal zone of acute ischemic cerebral apoplexy |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110909875.3A CN113808191B (en) | 2021-08-09 | 2021-08-09 | Automatic quantification and three-dimensional modeling method for focal zone of acute ischemic cerebral apoplexy |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113808191A true CN113808191A (en) | 2021-12-17 |
| CN113808191B CN113808191B (en) | 2024-06-11 |
Family
ID=78942897
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110909875.3A Active CN113808191B (en) | 2021-08-09 | 2021-08-09 | Automatic quantification and three-dimensional modeling method for focal zone of acute ischemic cerebral apoplexy |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113808191B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115311268A (en) * | 2022-10-10 | 2022-11-08 | 武汉楚精灵医疗科技有限公司 | Esophagus endoscope image identification method and device |
| CN115908457A (en) * | 2023-01-06 | 2023-04-04 | 脑玺(苏州)智能科技有限公司 | Low-density peduncle area segmentation method, analysis method, device, system, equipment and medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007058632A1 (en) * | 2005-11-21 | 2007-05-24 | Agency For Science, Technology And Research | Superimposing brain atlas images and brain images with delineation of infarct and penumbra for stroke diagnosis |
| CN101596109A (en) * | 2009-06-12 | 2009-12-09 | 深圳先进技术研究院 | Obtain method, system and the thrombolytic indication generation system and the method for brain characteristic parameter |
| CN108122221A (en) * | 2016-11-29 | 2018-06-05 | 中国科学院深圳先进技术研究院 | The dividing method and device of diffusion-weighted imaging image midbrain ischemic area |
| CN110111296A (en) * | 2019-01-30 | 2019-08-09 | 北京慧脑云计算有限公司 | The automatic segmenting system of small infarct lesion and its method under the new hair cortex of deep learning |
-
2021
- 2021-08-09 CN CN202110909875.3A patent/CN113808191B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007058632A1 (en) * | 2005-11-21 | 2007-05-24 | Agency For Science, Technology And Research | Superimposing brain atlas images and brain images with delineation of infarct and penumbra for stroke diagnosis |
| CN101596109A (en) * | 2009-06-12 | 2009-12-09 | 深圳先进技术研究院 | Obtain method, system and the thrombolytic indication generation system and the method for brain characteristic parameter |
| CN108122221A (en) * | 2016-11-29 | 2018-06-05 | 中国科学院深圳先进技术研究院 | The dividing method and device of diffusion-weighted imaging image midbrain ischemic area |
| CN110111296A (en) * | 2019-01-30 | 2019-08-09 | 北京慧脑云计算有限公司 | The automatic segmenting system of small infarct lesion and its method under the new hair cortex of deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 陈庆华;王向明;方权;李彦君;王振强;: "DWI在MCA区急性脑梗死患者诊断中的应用价值", 医学影像学杂志, no. 12 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115311268A (en) * | 2022-10-10 | 2022-11-08 | 武汉楚精灵医疗科技有限公司 | Esophagus endoscope image identification method and device |
| CN115908457A (en) * | 2023-01-06 | 2023-04-04 | 脑玺(苏州)智能科技有限公司 | Low-density peduncle area segmentation method, analysis method, device, system, equipment and medium |
| CN115908457B (en) * | 2023-01-06 | 2023-05-23 | 脑玺(苏州)智能科技有限公司 | Low-density infarct distinguishing method, analysis method, device, system, equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113808191B (en) | 2024-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ramzan et al. | Volumetric segmentation of brain regions from MRI scans using 3D convolutional neural networks | |
| Zhang et al. | MI-UNet: multi-inputs UNet incorporating brain parcellation for stroke lesion segmentation from T1-weighted magnetic resonance images | |
| Kuang et al. | Segmenting hemorrhagic and ischemic infarct simultaneously from follow-up non-contrast CT images in patients with acute ischemic stroke | |
| Saini et al. | Brain tumor detection in medical imaging using MATLAB | |
| Hu et al. | Automatic segmentation of intracerebral hemorrhage in CT images using encoder–decoder convolutional neural network | |
| EP2481349B1 (en) | Method, device, and program for processing images obtained from asl mri | |
| CN107545584A (en) | The method, apparatus and its system of area-of-interest are positioned in medical image | |
| CN103942799B (en) | Breast ultrasounography image segmentation method and system | |
| CN113808191A (en) | Automatic quantification and three-dimensional modeling method for focal zone of acute ischemic stroke | |
| Fan et al. | Lung nodule detection based on 3D convolutional neural networks | |
| Zheng et al. | MDCC-Net: multiscale double-channel convolution U-Net framework for colorectal tumor segmentation | |
| CN106709920B (en) | Blood vessel extraction method and device | |
| Kano et al. | Automatic contour segmentation of cervical cancer using artificial intelligence | |
| CN109949322A (en) | A kind of cerebrovascular image partition method based on magnetic resonance T1 enhancing image | |
| Rashed et al. | Development of accurate human head models for personalized electromagnetic dosimetry using deep learning | |
| CN110008992A (en) | A kind of deep learning method for prostate cancer auxiliary diagnosis | |
| CN113516653A (en) | Method for identifying glioma recurrence and necrosis by multi-feature fusion calculation | |
| CN112967786A (en) | Construction method and system of anatomical navigation based on multimode image and interactive equipment | |
| Çelik et al. | A new 3D MRI segmentation method based on Generative Adversarial Network and Atrous Convolution | |
| Xu et al. | RUnT: A network combining residual U-Net and transformer for vertebral edge feature fusion constrained spine CT image segmentation | |
| Zhao et al. | Research on 3D reconstruction algorithm of medical CT image based on parallel contour | |
| CN104484874B (en) | Living animal lower limb vascular dividing method based on CT contrast imagings | |
| Smith et al. | Automated torso contour extraction from clinical cardiac MR slices for 3D torso reconstruction | |
| Yu et al. | A benchmark dual-modality dental imaging dataset and a novel cognitively inspired pipeline for high-resolution dental point cloud synthesis | |
| Tomasetti et al. | Multi-input segmentation of damaged brain in acute ischemic stroke patients using slow fusion with skip connection |
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 |