Search

CN-121983290-A - General surgery department abdominal trauma grading evaluation method based on machine learning

CN121983290ACN 121983290 ACN121983290 ACN 121983290ACN-121983290-A

Abstract

The invention relates to the technical field of intelligent surgery, in particular to a general surgery abdominal trauma grading evaluation method based on machine learning, which comprises the following steps: according to the invention, the two-dimensional image is analyzed into a three-dimensional voxel matrix, a vascular skeleton and focus voxels are extracted, topological coherent features between the blood vessel and the lesion kitchen are captured, the spatial affinity of damaged tissues to a vascular network is quantified, a liquid dispersion area conforming to physical permeation features is screened, the survival potential of the tissues is evaluated according to the compensatory perfusion proportion of the isolated area, a theoretical critical state vector containing multidimensional physiological features is constructed, the severity of the wound is converted into a measurable spatial deviation value, and the precise quantitative grading from the anatomical morphology to functional compensatory dimension is realized.

Inventors

  • ZHOU DONGMEI
  • WANG ZHI
  • RAO YUN

Assignees

  • 中国人民解放军陆军军医大学第一附属医院

Dates

Publication Date
20260505
Application Date
20260211

Claims (10)

  1. 1. The general surgery department abdominal trauma grading evaluation method based on machine learning is characterized by comprising the following steps of: s1, acquiring an abdomen enhancement tomographic image, analyzing the abdomen enhancement tomographic image into a three-dimensional voxel matrix, dividing the three-dimensional voxel matrix into a high-density voxel set, a low-density voxel set and a background voxel set according to density distribution characteristics, constructing a portal vein skeleton and an arterial skeleton according to the background voxel set, and outputting anatomical structure characteristic data; s2, calling the anatomical structure feature data, extracting a low-density voxel set and a vascular skeleton, transforming the distances between a portal vein skeleton and an arterial skeleton, mapping the distances to the low-density voxel set, establishing a simplex complex filtering flow according to the distances, counting the number of the continuous features, and generating a vascular affinity topological vector; S3, calling the anatomical structure feature data, calculating a high-density voxel set edge gradient vector and a module value sequence attenuation rate along the gradient vector, screening a voxel set which is linearly and gently attenuated and has no mutation, calculating the volume, and outputting a liquid dispersoid volume parameter; S4, calling the anatomical structure feature data, identifying a portal vein interruption node and a downstream broken voxel set which are cut off by a low-density voxel set, counting the number of the broken voxel sets which are positioned in the arterial skeleton perfusion radius, calculating the duty ratio, and outputting the compensatory perfusion ratio; and S5, normalizing the vessel affinity topological vector, the liquid dispersion volume parameter and the compensatory perfusion ratio, combining the vessel affinity topological vector, the liquid dispersion volume parameter and the compensatory perfusion ratio into a multi-dimensional feature vector, constructing a critical state vector, calculating the weighted distance between the critical state vector and the multi-dimensional feature vector, matching a grading interval, and outputting an abdomen wound grading evaluation result.
  2. 2. The machine learning based general surgery abdominal trauma grading assessment method according to claim 1, wherein the anatomical feature data comprises a three-dimensional voxel matrix, voxel set partitioning results and skeleton construction results, the vessel affinity topology vector comprises a near vessel connected component number, a near vessel topology loop number and a topology feature duration interval length, the liquid dispersion volume parameter comprises a liquid penetration region volume value, a penetration region edge gradient continuity index and a penetration region average density value, the compensatory perfusion ratio comprises a portal vein injury region total voxel amount, an arterial compensatory coverage region voxel amount and a functional retention percentage value, and the abdominal trauma grading assessment results comprise a trauma severity grading label, a multi-dimensional feature weighted composite score and a functional retention map data.
  3. 3. The machine learning based general surgery abdominal trauma grading evaluation method according to claim 1, wherein the step of acquiring the anatomical feature data specifically comprises: S101, acquiring an abdomen enhancement tomographic image, analyzing the abdomen enhancement tomographic image into a three-dimensional voxel matrix, analyzing the gray level intensity distribution characteristics of voxel units in the matrix, classifying and screening according to the density difference of various tissues in a contrast state, dividing the three-dimensional voxel matrix into a high-density voxel set, a low-density voxel set and a background voxel set, and generating a basic voxel division set; S102, calling the basic voxel segmentation set, extracting a background voxel set, carrying out morphological refinement contraction treatment on the background voxel set, stripping non-core tissue pixels, reserving a communication path with single pixel width, calculating the geometric center line coordinate of a residual path, tracking and classifying the center line according to a vessel anatomical connectivity rule, constructing a portal vein communication network and an arterial communication network, and generating vessel skeleton data; S103, calling a high-density voxel set and a low-density voxel set in the basic voxel segmentation set, combining the blood vessel skeleton data, establishing a three-dimensional space mapping coordinate system, mapping the space distribution coordinates of each voxel set to the coordinate system, carrying out space position registration by taking the blood vessel skeleton data as a reference, and fusing the physical density attribute and skeleton topological structure information of each group of voxels to generate anatomical structure feature data.
  4. 4. The machine learning based general surgery abdominal trauma grading evaluation method according to claim 3, wherein the classifying and screening process according to the density difference of various tissues in the contrast state, the process of dividing the three-dimensional voxel matrix into a high-density voxel set, a low-density voxel set and a background voxel set comprises the following specific steps: counting the gray level intensity values of all voxel units in the three-dimensional voxel matrix, and establishing a global gray level statistical histogram reflecting the overall distribution form of the abdomen tissue density; Carrying out probability density function fitting on the global gray statistical histogram by using a Gaussian mixture model, analyzing three independent Gaussian distribution components, and calculating the gray average value of each Gaussian distribution component; Arranging the three Gaussian distribution components according to the sequence from small to large of the gray average value, and sequentially defining the three Gaussian distribution components as a low attenuation distribution component, a background distribution component and a high attenuation distribution component; Calculating the intersection point gray level value of the probability density curve between the low attenuation distribution component and the background distribution component, and setting the intersection point gray level value as a low density segmentation limit value; Calculating the intersection point gray level value of the probability density curve between the background distribution component and the high attenuation distribution component, and setting the intersection point gray level value as a high density segmentation limit value; Traversing the voxel units in the three-dimensional voxel matrix, and extracting the voxel units with gray level intensity values larger than the high-density segmentation limit value to a high-density voxel set; extracting voxel units with gray level intensity values smaller than the low-density segmentation limit value to a low-density voxel set; And extracting voxel units with gray level intensity values within the low-density segmentation limit value and the high-density segmentation limit value closed interval to a background voxel set.
  5. 5. The machine learning based general surgery abdominal trauma grading evaluation method according to claim 4, wherein the vessel affinity topology vector acquisition step specifically comprises: s201, calling the anatomical structure feature data, extracting a low-density voxel set and a blood vessel skeleton, establishing a three-dimensional air space distance field by taking the blood vessel skeleton as a reference standard, calculating Euclidean distance parameters from full-field voxels to nearest blood vessel skeleton points, traversing voxel units in the low-density voxel set, searching corresponding position coordinates of each voxel unit in the distance field, and endowing the Euclidean distance parameters to the corresponding low-density voxels to generate a blood vessel distance attribute mapping set; S202, invoking the vascular distance attribute mapping set, sequencing voxels in the mapping set according to the increasing sequence of distance parameters, constructing a simplex complex evolution sequence, analyzing the morphological changes of connected components and topology holes in the sequence evolution process, recording the generation time and the extinction time of each topological structure in the simplex complex, acquiring the persistence characteristics of the lacerated structure along with the vascular distance change, and generating topology characteristic persistence interval data; S203, invoking the topological feature duration interval data, screening and generating topological features with the moment in the initial parameter range, classifying and counting the continuous number of the same class of different dimensions in the target range, analyzing the distribution density condition of the topological features on the distance parameter axis, arranging and combining the counted number indexes and the distribution density indexes according to a preset sequence, constructing a digitalized vector, and generating a vessel affinity topological vector.
  6. 6. The machine learning based general surgery abdominal trauma grading evaluation method according to claim 5, wherein the step of obtaining the fluid dispersion volume parameter specifically comprises: S301, calling the anatomic structure feature data, extracting a high-density voxel set, screening edge voxels at the junction of high-density tissues and a background in the set, calculating the gray level change gradient of the edge voxels in a local three-dimensional environment, constructing vector data containing the maximum change rate direction and gradient strength information, and generating edge gradient vector field data; S302, calling the data of the edge gradient vector field, analyzing the directional distribution condition of a gradient vector in a three-dimensional space, collecting the gradient strength of continuous neighborhood voxels along the extending direction of the gradient vector, constructing a strength change sequence distributed along a space distance, calculating the differential attenuation rate of the gradient strength relative to the space position in the sequence, and generating a space attenuation rate sequence; S303, calling the spatial attenuation rate sequence, analyzing the evolution rule of the gradient strength on the spatial extension path, screening a voxel set with the gradient modulus value which is linearly and gently attenuated along with the distance and has no step mutation, taking the voxel set as a liquid dispersion characteristic region, calculating the cumulative space occupation amount, and generating a liquid dispersoid volume parameter.
  7. 7. The machine learning based general surgery abdominal trauma grading evaluation method according to claim 6, wherein the process of screening the voxel set with linear gradual attenuation of gradient modulus value along with distance and without step mutation as the liquid dispersion characteristic region specifically comprises: Performing least square linear regression fitting on discrete gradient modulus data contained in the space attenuation rate sequence, and calculating a linear correlation coefficient representing the overall change trend of the sequence and a slope value of a fitting straight line; Calculating the differential absolute value of the gradient modulus value of the adjacent space position in the space attenuation rate sequence point by point, and constructing a first-order differential sequence reflecting the local numerical jump characteristic; Counting the arithmetic average value of all voxel unit gradient modulus values in the background voxel set, setting the arithmetic average value as a gentle attenuation reference standard, and setting the preset lower limit of the linear correlation coefficient as a linear correlation threshold; Calculating the average value and standard deviation of the absolute value of the difference in the sequence based on the numerical distribution characteristics of the first-order difference sequence, and setting the superposition sum of the average value and three times of the standard deviation as a dynamic step judgment threshold value; screening the sequence of which the linear correlation coefficient is larger than the linear correlation threshold value, the slope value is a negative value and the absolute value is smaller than the gentle decay reference standard, and judging that the linear gentle decay condition is met; Traversing the first-order differential sequence corresponding to the sequence meeting the linear gentle attenuation condition, checking whether a numerical value point larger than the dynamic step judgment threshold exists in the sequence, and eliminating the sequence with the numerical value point; And extracting and marking the voxel unit set which is screened by the linear gentle attenuation condition and corresponds to the sequence without the numerical point as a liquid dispersion characteristic region.
  8. 8. The machine learning based general surgery abdominal trauma grading evaluation method according to claim 7, wherein the compensating perfusion ratio obtaining step specifically comprises: S401, calling the anatomical structure feature data, extracting a low-density voxel set and a vascular skeleton, establishing a three-dimensional space superposition mapping relation of the low-density voxel set and the portal vein vascular skeleton, analyzing the path communication state of the vascular skeleton in a low-density region, judging the space geometric position of the physical interception of the skeleton path, marking a breakpoint lacking upstream connectivity and taking the breakpoint as a blood flow transmission interruption starting point, and generating a portal vein interruption node set; s402, invoking the portal vein interrupt node set, performing downstream path traversing operation according to the topological extension direction of a vascular anatomical structure, searching a vascular branch network distribution area downstream of the interrupt node, collecting all voxel units in the extension range of a branch network, constructing damaged tissue space distribution data losing portal vein connectivity, and generating a downstream interrupt voxel set; S403, calling the downstream broken voxel set, correlating arterial skeletons in the anatomical structure characteristic data, calculating the space proximity between the broken voxels and the nearest arterial skeleton, screening voxel units positioned in the arterial blood permeation coverage range, and counting the proportional relation of the size of the screened voxels to the total size of the broken voxel set to generate a compensatory perfusion ratio.
  9. 9. The machine learning based general surgery abdominal trauma grading evaluation method according to claim 8, wherein the step of obtaining the abdominal trauma grading evaluation result specifically comprises: S501, acquiring the vessel affinity topological vector, the liquid dispersion volume parameter and the compensatory perfusion ratio, analyzing physical dimension differences and data distribution range characteristics, carrying out normalized mapping treatment, and establishing a unified high-dimensional characteristic space coordinate system by the multi-source characteristic data after fusion treatment to generate a multi-dimensional characteristic vector; S502, calling the multidimensional feature vector, constructing a critical reference standard composed of theoretical limit states of all evaluation dimensions, establishing a hierarchical evaluation interval sequence containing multiple severity levels, analyzing the contribution weight difference of each feature dimension in a wound severity evaluation system, calculating the weighted Euclidean distance between the current feature vector and the critical reference standard in a multidimensional feature space, and generating distance difference calculation data; And S503, searching the corresponding interval of the distance difference calculation data according to the grading evaluation interval sequence, judging the severity level category of the current wound state, and generating an abdomen wound grading evaluation result.
  10. 10. The machine learning based general surgery abdominal trauma grading evaluation method according to claim 9, wherein the process of constructing a critical reference standard composed of theoretical limit states of each evaluation dimension and establishing a grading evaluation interval sequence containing multiple severity grades is specifically as follows: Judging the maximum theoretical value representing the complete coincidence of the fracture and the vascular skeleton space in the vascular affinity topological vector, and defining the maximum theoretical value as a topological erosion limit value; Judging the numerical value of the maximum physical accommodation quantity of the abdominal cavity anatomical gap represented by the liquid dispersion volume parameter, and defining the numerical value as a dispersion capacity limit value; Judging a zero value representing the complete uncompensated effect of the artery in the compensatory perfusion ratio, and defining a perfusion deficiency limit value; Performing normalization processing on the topological erosion limit value, the dispersion capacity limit value and the perfusion deficiency limit value, and constructing a critical reference standard in a multidimensional feature space in a combined manner; The classified historical abdominal trauma case data are called, the weighted Euclidean distance between the multidimensional feature vector of each historical case and the critical reference standard is calculated, and a sample distance set is constructed; Performing iterative operation on the sample distance set by using a K-means clustering algorithm to obtain clustering center values corresponding to different wound severity degrees, and sequencing the clustering center values according to the sequence from small values to large values; calculating the arithmetic average value of the numerical values of two adjacent clustering centers, setting the arithmetic average value as a grading boundary threshold value, dividing a continuous numerical range based on the grading boundary threshold value, and establishing a grading evaluation interval sequence.

Description

General surgery department abdominal trauma grading evaluation method based on machine learning Technical Field The invention relates to the technical field of intelligent surgery, in particular to a general surgery abdominal trauma grading evaluation method based on machine learning. Background The technical field of intelligent surgery covers a medical practice system which integrates artificial intelligence, computer vision, robot technology and modern surgery depth, and relates to the digital reconstruction and analysis of multi-mode medical images by utilizing computer assistance, the fine operation of a surgical robot or an optical navigation assisting doctor, the quantitative analysis of surgical risks, focus characteristics and prognosis conditions by utilizing an algorithm model, and the construction of a full-flow digital diagnosis and treatment environment which covers preoperative planning, intraoperative real-time assistance and postoperative rehabilitation assessment. The traditional general surgery abdomen trauma grading evaluation method is a diagnosis and treatment process of carrying out clinical grading on the injury severity of an abdomen parenchymal organ or a cavity organ after external force impact, a clinician mainly carries out comprehensive research and judgment according to an abdomen enhanced electronic computer tomography image of a patient, by observing and identifying the organ laceration depth, the area ratio of hematoma surface under an envelope, the diameter of the parenchymal hematoma and whether active bleeding contrast agent overflows or not through naked eyes, manually measuring anatomical geometric parameters of an injury area on an image workstation by using a caliper tool in combination with a trauma organ injury grading standard, and then checking by item by comparison with a grading table to determine specific injury grades, and synchronously combining the contraction pressure, pulse frequency and abdomen physical examination sign data of the patient to finish the definition of the critical degree of illness. The traditional manual evaluation mainly relies on two-dimensional tomographic images for visual study and judgment, three-dimensional non-Euclidean space morphology of organ injury cannot be accurately restored, the irregular wound volume is difficult to quantify by measuring the fracture depth or hematoma diameter on a single layer by a caliper tool, key anatomical details are omitted due to the fact that manual subjective reading is easily influenced by experience differences, topological connectivity between a vascular skeleton and damaged tissues is omitted by simple geometric morphology measurement, compensatory perfusion functions of vessels and microscopic dispersion behaviors of liquid media in tissue gaps are difficult to dynamically reflect by static anatomical parameters, pathological critical states of complex and complicated injury conditions cannot be accurately defined by stiff table lookup classification, and hysteresis deviation exists in quantitative rating of disease severity. Disclosure of Invention In order to achieve the above purpose, the invention adopts the following technical scheme that the general surgery department abdominal trauma grading evaluation method based on machine learning comprises the following steps: s1, acquiring an abdomen enhancement tomographic image, analyzing the abdomen enhancement tomographic image into a three-dimensional voxel matrix, dividing the three-dimensional voxel matrix into a high-density voxel set, a low-density voxel set and a background voxel set according to density distribution characteristics, constructing a portal vein skeleton and an arterial skeleton according to the background voxel set, and outputting anatomical structure characteristic data; s2, calling the anatomical structure feature data, extracting a low-density voxel set and a vascular skeleton, transforming the distances between a portal vein skeleton and an arterial skeleton, mapping the distances to the low-density voxel set, establishing a simplex complex filtering flow according to the distances, counting the number of the continuous features, and generating a vascular affinity topological vector; S3, calling the anatomical structure feature data, calculating a high-density voxel set edge gradient vector and a module value sequence attenuation rate along the gradient vector, screening a voxel set which is linearly and gently attenuated and has no mutation, calculating the volume, and outputting a liquid dispersoid volume parameter; S4, calling the anatomical structure feature data, identifying a portal vein interruption node and a downstream broken voxel set which are cut off by a low-density voxel set, counting the number of the broken voxel sets which are positioned in the arterial skeleton perfusion radius, calculating the duty ratio, and outputting the compensatory perfusion ratio; and S5, normalizing the vessel affinity to