Multi-dimensional detection method based on minimal lesion in children kidney CT image

Abstract

The invention relates to the technical field of detection of minimal lesions, and discloses a multi-dimensional detection method based on minimal lesions in a children kidney CT (Computed Tomography) image. Constructing a standard comprehensive quality evaluation model and a dynamic age correction model according to the calculated multi-dimensional parameters of the minimal lesions in the kidney CT image, obtaining a standard score threshold value, and taking the standard score threshold value as an evaluation standard; then, carrying out quantitative analysis on the uncertainty of the standard comprehensive quality evaluation model, obtaining a quality evaluation matrix and a real-time comprehensive quality evaluation model, and obtaining a real-time comprehensive quality score, namely a single detection result, by utilizing the real-time comprehensive quality evaluation model; and finally, the real-time comprehensive quality score is compared with a standard score threshold, accurate identification of minimal lesions in the children kidney CT image is realized, and the technical problems of age difference and insufficient consensus mechanism in the identification process are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of the detection of minute lesions, in particular to a multi-dimensional detection method for minute lesions in pediatric kidney CT images. Background Art

[0002] Dynamic contrast-enhanced examination of the kidney is an important technical means for identifying hidden small lesions in the kidney. The CT imaging examination of the kidney generates a large amount of image data. Manual screening of images is laborious and inefficient. Due to the diversity and complexity of kidney lesions, the quantification of many indicators and the evaluation of pathological grading rely on manual counting and semi-quantitative evaluation. The whole evaluation process is time-consuming and laborious, and has subjective differences. Moreover, manual visual evaluation will ignore some minute lesions, resulting in problems such as missed diagnosis.

[0003] In order to solve the above technical problems, those skilled in the art have disclosed a detection method for minute lesions in pediatric kidney CT images, and its publication number is: CN116167966A; this method includes the annotation of minute lesion foci in pediatric kidney CT images and the construction of a database; the extraction of features of minute lesion foci in pediatric kidney CT images based on a dynamic weighted skip network; the localization of minute lesion foci in pediatric kidney CT images based on a non-maximum suppression strategy; the dynamic weighted skip network outputs the foci results; thus, through the designed dynamic weighted skip network, the automatic extraction of foci features is realized, and through the designed non-maximum suppression strategy, the accurate identification of minute lesion foci in kidney CT images in pediatric kidney medical imaging data is realized.

[0004] However, the following problems still exist in the actual implementation of the above patent:

[0005] 1. Ignoring age differences: The 300% developmental differences in the kidney volume of children aged 1-14 are not considered, resulting in inaccurate evaluation criteria and unable to meet the precise detection needs of children of different ages;

[0006] 2. Insufficient consensus mechanism: Due to the large differences in the subjective recognition of pediatric kidney CT images, the recognition of a single doctor has certain limitations. There is a lack of mathematical modeling for the differences in multi-expert annotations in its method, and it is difficult to identify abnormal annotations;

[0007] 3. Lack of evaluation and verification capabilities: Since the recognition results of minute lesions in pediatric kidney CT images are easily affected by various subjective and objective factors, the obtained detection results in real time have great uncertainties and cannot stably and reliably reflect the lesion conditions, lacking the ability to verify and evaluate the detection results.

[0008] Therefore, those skilled in the art need a detection method that can perform multi-dimensional analysis on the micro-lesions in children's kidney CT images, so as to improve the detection rate of kidney micro-lesions and reduce the missed diagnosis rate of kidney micro-lesions. Summary of the Invention

[0009] The purpose of the present invention is to solve the problems of poor reliability and obvious differences in the existing detection methods for micro-lesions in children's kidney CT images, and a multi-dimensional detection method based on micro-lesions in children's kidney CT images is designed.

[0010] To achieve the above object, the technical solution of the present invention is a multi-dimensional detection method based on micro-lesions in children's kidney CT images, and the method includes the following steps:

[0011] Step 1, construct a children's kidney anatomical template database, and group the children's kidney anatomical templates in the database according to the age of the children;

[0012] Step 2, calculate the spatial anatomical fitness, morphological rationality index, expert consensus degree, and clinical consistency factor of the micro-lesions in the kidney CT image based on the children's kidney anatomical template database, and construct a standard comprehensive quality assessment model, which will be used as the basis for lesion detection and result evaluation;

[0013] Step 3, introduce an age correction function to perform non-linear calibration on the standard comprehensive quality assessment model to obtain a dynamic age correction model, and use the dynamic age correction model to output a standard score threshold, and the main function of this accurate score threshold is to be used as a standard for judging the reliability of the detection result;

[0014] Step 4, perform a quantitative analysis on the uncertainty of the standard comprehensive quality assessment model to obtain a quality assessment matrix, and then use the quality assessment matrix to construct a real-time comprehensive quality assessment model and obtain a real-time comprehensive quality score, and the real-time comprehensive quality score will be used as the single detection result of the micro-lesions in the children's kidney CT image;

[0015] Step 5, compare the real-time comprehensive quality score with the standard score threshold to evaluate the reliability of the real-time comprehensive quality score through the comparison. If the real-time comprehensive quality score is reliable, the real-time comprehensive quality score will be used as the result output of the detection of the micro-lesions in the children's kidney CT image; if the real-time comprehensive quality score is not reliable, the real-time comprehensive quality score will be cancelled and the operation of Step 4 will be repeated until a reliable real-time comprehensive quality score is obtained; through the above comparison operation, accurate identification of the micro-lesions in the children's kidney CT image can be achieved.

[0016] In the Step 1, the children's kidney anatomical template database includes, but is not limited to: annotation masks, anatomical templates; the age of the children does not exceed fourteen years old.

[0017] The spatial anatomical fitness S in the second step C has the following calculation formula:

[0018]

[0019] In the formula, M is the labeled mask, T is the age-matched anatomical template, C M , C T are the labeled coordinates and the centroid coordinates of the template, D T is the maximum diameter of the template kidney, ω1 = 0.6, ω2 = 0.4;

[0020] The morphological rationality index M P has the following calculation formula:

[0021]

[0022] In the formula, A is the area of the labeled region, P is the perimeter of the labeled region, σ r / μ r is the ratio of the standard deviation to the mean in the radial direction, S k is the Hausdorff distance similarity with the standard ellipsoid, and Sigmoid is a mathematical transformation function;

[0023] The expert consensus degree E C has the following calculation formula:

[0024]

[0025] In the formula, p i is the labeled probability map of the i-th expert, D KL is the Kullback-Leibler divergence, and N is the number of pediatric imaging diagnosis experts participating in the labeling and N > 3;

[0026] The clinical consistency factor C f has the following calculation formula:

[0027]

[0028] In the formula, TP is the number of true tiny lesions correctly identified in the labeling result; FP is the number of normal tissue regions mislabeled as lesions in the labeling result; FN is the number of tiny lesions that actually exist but are not labeled; and λ1 is the miss penalty coefficient.

[0029] The standard comprehensive quality assessment model in the second step is:

[0030]

[0031] Wherein, Q is the standard comprehensive quality score, and α, β, and γ are all weight coefficients, and the weight coefficients can be dynamically calibrated through the Bayesian optimization algorithm.

[0032] The dynamic calibration model of the weight coefficient based on the Bayesian optimization algorithm is:

[0033]

[0034] Wherein, θ is the parameter to be optimized, θ = (α, β, γ); θ * is the optimized parameter, Q gold is the gold standard score determined by the expert group, Q is the standard comprehensive quality score, and λ is the adjustment coefficient.

[0035] The dynamic age correction model in the third step is:

[0036]

[0037] Wherein, Q final is the standard score threshold, and age is the age of the child to be detected.

[0038] The quantitative analysis formula for uncertainty in the fourth step is:

[0039]

[0040] Wherein, U Q is the quantitative analysis value of uncertainty, ∈ is to prevent the zero factor, ∈ = 0.01 μ is the historical mean value, μ s is the historical mean value of the spatial anatomical fitness, μ M is the historical mean value of the morphological rationality index, μ E is the historical mean value of the expert consensus degree.

[0041] The quality assessment matrix QM in the fourth step is:

[0042]

[0043] The real-time comprehensive quality assessment model in the fourth step is:

[0044] Q = ||QM|| F ·σ max (QM) (10)

[0045] Wherein, ||·|| F is the Frobenius norm, and σ max is the largest singular value.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. This method constructs a standard model for evaluation and a real-time model for detection based on multi-dimensional influencing factors. At the same time, it fully considers the differences in kidney development among children of different ages, introduces the age mechanism of children into the detection and evaluation process, effectively controls the misalignment rate of detection results, and thus better meets the accurate detection needs of children of different ages.

[0048] 2. This method has certain limitations in the identification of minor diseases by a single expert. At the same time, it fully considers the subjective identification differences of different experts in the minor lesions in children's kidney CT images, constructs a mathematical model that meets multi-expert annotation and recognition, and thus better meets the accurate detection needs of minor diseases.

[0049] 3. This method effectively controls the influence of subjective and objective factors on the detection results through comparative analysis, so as to stably and reliably reflect the lesion situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flowchart of a multi-dimensional detection method for minor lesions in children's kidney CT images according to the present invention;

[0051] Figure 2 is a system framework diagram of a multi-dimensional detection method for minor lesions in children's kidney CT images according to the present invention;

[0052] Figure 3 is a comparative analysis table of the detection effects of a multi-dimensional detection method for minor lesions in children's kidney CT images according to the present invention and the detection effects of existing minor lesion detection methods;

[0053] Figure 4 is a data table for age correction of children's quality scores (1 - 14 years old) according to the present invention;

[0054] Figure 5 is a comparative table for evaluating the quality of annotation of children's kidney CT images according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0055] The present invention will be specifically described below with reference to the accompanying drawings, as Figures 1-5 shown;

[0056] A multi-dimensional detection method for minor lesions in children's kidney CT images, the method comprising the following steps:

[0057] Step 1, construct a children's kidney anatomical template database, and group the children's kidney anatomical templates in the database according to the age of the children; the process is as follows: collect 500 children's kidney CT images, construct a children's kidney anatomical template database by dividing them into 4 groups according to age, and the annotation data in the database includes the independent annotation results of 3 senior physicians.

[0058] Step 2: Calculate the spatial anatomical fitness, morphological rationality index, expert consensus degree, and clinical consistency factor of the micro-lesions in the pediatric kidney CT images based on the pediatric kidney anatomical template database. Then, construct a standard comprehensive quality assessment model for the micro-lesions in the pediatric kidney CT images based on the spatial anatomical fitness, morphological rationality index, expert consensus degree, and clinical consistency factor.

[0059] Step 3: Introduce an age correction function to perform non-linear calibration on the standard comprehensive quality assessment model and obtain a dynamic age correction model. Use the dynamic age correction model to output the standard scoring threshold for the micro-lesions in the pediatric kidney CT images.

[0060] Step 4: Quantitatively analyze the uncertainty of the standard comprehensive quality assessment model. Based on the results of the uncertainty quantitative analysis of the standard comprehensive quality assessment model, construct a quality assessment matrix. Then, use the quality assessment matrix to construct a real-time comprehensive quality assessment model and obtain a real-time comprehensive quality score.

[0061] Step 5: Compare the real-time comprehensive quality score with the standard scoring threshold. When the real-time comprehensive quality score is greater than or equal to the standard scoring threshold, the recognition result of the micro-lesions in the pediatric kidney CT images is reliable; when the real-time comprehensive quality score is less than the standard scoring threshold, the recognition result of the micro-lesions in the pediatric kidney CT images is unreliable, thus achieving the accurate recognition of the micro-lesions in the pediatric kidney CT images.

[0062] The creative point of this application lies in constructing a standard comprehensive quality assessment model by integrating multi-dimensional factors such as the spatial anatomical fitness, morphological rationality index, expert consensus degree, and clinical consistency factor of the pediatric kidney anatomical template. Using the standard comprehensive quality assessment model as the basis for lesion detection and evaluation, optimize the standard comprehensive quality assessment model by fully considering the specificity of the pediatric age and obtain the standard scoring threshold for evaluation. The standard scoring threshold will be used as the standard for evaluating the detection results to analyze the reliability of each detection result; then, perform uncertainty analysis on the standard comprehensive quality assessment model to obtain the real-time comprehensive quality score. This real-time comprehensive quality score will be used as the result of the detection of the micro-lesions in the pediatric kidney CT images, and this result is only the single detection result of the micro-lesions in the pediatric kidney CT images. Finally, evaluate the reliability of the single detection result through comparative analysis, thereby ensuring the accuracy of the detection of the micro-lesions in the pediatric kidney CT images. The comparative analysis of the detection effects of the present invention and the existing micro-lesion detection methods is as Figure 3 shown.

[0063] In the technical solution of this application, the pediatric kidney anatomical template database includes, but is not limited to: annotation masks and anatomical templates; and the age of the child does not exceed fourteen years old.

[0064] It should be noted that the calculation of the spatial anatomical fitness S C is given by the formula:

[0065]

[0066] where M is the annotation mask, T is the age-matched anatomical template, C M ,C T are the annotation coordinates and the centroid coordinates of the template, D T is the maximum diameter of the template kidney, ω1 = 0.6, ω2 = 0.4;

[0067] The calculation of the morphological rationality index M P is given by the formula:

[0068]

[0069] where A is the area of the annotation region, P is the perimeter of the annotation region, σ r / μ r is the ratio of the standard deviation in the radial direction to the mean, S k is the Hausdorff distance similarity to the standard ellipsoid, and Sigmoid is a mathematical transformation function;

[0070] The calculation of the expert consensus degree E C is given by the formula:

[0071]

[0072] where p i is the annotation probability map of the i-th expert, D KL is the Kullback-Leibler divergence, and N is the number of pediatric imaging diagnosis experts participating in the annotation and N > 3;

[0073] The calculation of the clinical consistency factor C f is given by the formula:

[0074]

[0075] Wherein, TP (True Positives) is the number of truly identified minor lesions in the annotation results. Its calculation method is to compare with the gold standard (pathological biopsy or expert confirmation), and the part with a spatial overlap degree (such as DICE coefficient ≥ 0.7) between the annotation area and the true lesion and consistent pathological features. It is mainly used to reflect the algorithm's ability to capture true lesions; FP (False Positives) is the number of normal tissue areas mislabeled as lesions in the annotation results. Its calculation method is the part where the annotation area does not overlap with the gold standard lesion (DICE coefficient < 0.1) and the pathological verification is negative. Its main role is to measure the risk of overdiagnosis of the algorithm; FN (False Negatives) is the number of minor lesions that actually exist but are not annotated. Its calculation method is the part where the lesions confirmed by the gold standard are not covered in the annotation results (DICE coefficient < 0.1). Its main role is to reflect the possibility of missed diagnosis of the algorithm; λ1 is the missed diagnosis penalty coefficient, which is a weight coefficient used to adjust the influence of false negatives on the score. λ1 = 0.3. The harm of missed diagnosis in pediatric kidney diseases is relatively high (such as delaying treatment), and it needs to be punished but not overly reduce the score sensitivity. It is determined by ROC curve analysis. When λ = 0.3, the AUC value of C f reaches 0.91 (optimal balance point); its main role is when λ < 1, to weaken the contribution of missed diagnosis to the denominator, avoid the score being overly sensitive to the number of missed diagnoses, and ensure that the score achieves a balance between the detection rate and the false positive rate.

[0076] In addition, the constructed standard comprehensive quality assessment model is:

[0077]

[0078] Wherein, Q is the standard comprehensive quality score, and α, β, and γ are all weight coefficients. The weight coefficients can be dynamically calibrated through the Bayesian optimization algorithm, and the values of the weight coefficients are: α = 0.45, β = 0.3, γ = 0.25.

[0079] Among them, the dynamic calibration model of the weight coefficients based on the Bayesian optimization algorithm is:

[0080]

[0081] Wherein, θ is the parameter to be optimized, θ = (α, β, γ); θ * is the optimized parameter, Q gold is the gold standard score determined by the expert group, Q is the standard comprehensive quality score, and λ is the adjustment coefficient.

[0082] And the constructed dynamic age correction model is:

[0083]

[0084] Wherein, Q final is the standard scoring threshold, and age is the age of the child to be detected.

[0085] In the technical solution of the present application, the formula for quantitative analysis of uncertainty is:

[0086]

[0087] Wherein, U Q is the quantitative analysis value of uncertainty, ∈ is to prevent zero factors, ∈ = 0.01, μ is the historical mean, μ s is the historical mean of spatial anatomical fitness, μ M is the historical mean of the morphological rationality index, μ E is the historical mean of the expert consensus degree.

[0088] By comprehensively considering the quantitative analysis results of uncertainty and the calculation results of spatial anatomical fitness, morphological rationality index, expert consensus degree, and clinical consistency factor, the quality assessment matrix QM can be obtained as:

[0089]

[0090] Then, based on the quality assessment matrix QM, the real-time comprehensive quality assessment model can be obtained as:

[0091] Q = ||QM|| F ·σ max (QM)(10)

[0092] Wherein, ||·|| F is the Frobenius norm, and σ max is the maximum singular value.

[0093] Example 1:

[0094] Based on the PACS system of a certain hospital, the system framework of the present method is deployed, which includes: a data input layer, a multi-dimensional analysis engine, a dynamic correction module, and a visualization output layer, as Figure 2 shown;

[0095] Among them,

[0096] The data input layer can access the PACS system of the hospital and obtain the permissions for data acquisition and communication. Its main function is to process the acquired data into multi-dimensional data (such as: spatial anatomical fitness, morphological rationality index, expert consensus degree, and clinical consistency factor, etc.) for output;

[0097] The function of the multi-dimensional analysis engine is to receive the multi-dimensional data output by the data input layer, analyze the received multi-dimensional data, and then output a real-time comprehensive quality score;

[0098] The function of the dynamic correction module is to receive the real-time comprehensive quality score output by the multi-dimensional analysis engine and the multi-dimensional data output by the data input layer. The dynamic correction module can generate a standard score threshold based on the multi-dimensional data, compare the real-time comprehensive quality score with the standard score threshold, and output a reliable real-time comprehensive quality score as the detection result according to the comparison result;

[0099] The visualization output layer can receive the detection results output by the dynamic correction module. Its main function is to visually output the received detection results. The objects of visual output include but are not limited to: computers, smartphones, and tablet computers.

[0100] It should be noted that Figure 4 the basic score threshold of the child quality score age correction data table in: Q = 0.82 (determined by the ROC curve), the peak correction coefficient reaches the maximum value of 1.05 at the age of 7 (corresponding to the sine function phase angle π / 2), and the non-linear mapping reflects the growth law of the child's kidney volume (the number of glomeruli increases by 240% from 1 to 7 years old).

[0101] In addition, Figure 5 the conditions for the quality assessment of child kidney CT image annotation in include: the sample size is 200 cases (including 100 cases of <5mm micro-lesions); the equipment uses GE Revolution CT vs Siemens Somatom Force; and the gold standard uses ultrasound-guided percutaneous biopsy.

[0102] The above technical solutions only reflect the preferred technical solutions of the technical solutions of the present invention. Some changes that those skilled in the art may make to some parts thereof all reflect the principles of the present invention and fall within the protection scope of the present invention.

Claims

1. A multi-dimensional detection method for minute lesions in children's kidney CT images, characterized in that The method includes the following steps: Step 1: Construct a database of pediatric kidney anatomical templates and group the pediatric kidney anatomical templates in the database according to the age of the children; Step 2: Calculate the spatial anatomical fitness, morphological rationality index, expert consensus degree, and clinical consistency factor of micro-lesions in kidney CT images based on the database of pediatric kidney anatomical templates, and construct a standard comprehensive quality assessment model; Step 3: Introduce an age correction function to non-linearly calibrate the standard comprehensive quality assessment model to obtain a dynamic age correction model, and use the dynamic age correction model to output a standard score threshold; Step 4: Quantitatively analyze the uncertainty of the standard comprehensive quality assessment model to obtain a quality assessment matrix, and then use the quality assessment matrix to construct a real-time comprehensive quality assessment model and obtain a real-time comprehensive quality score; Step 5: Compare the real-time comprehensive quality score with the standard score threshold to achieve accurate identification of micro-lesions in pediatric kidney CT images.

2. The multi-dimensional detection method for minute lesions in children's kidney CT images according to claim 1, wherein, The database of pediatric kidney anatomical templates in Step 1 includes, but is not limited to, labeled masks and pediatric kidney anatomical templates.

3. A multi-dimensional detection method for minute lesions in children's kidney CT images according to claim 1, characterized in that The spatial anatomical adaptability S in the second step C is calculated by the following formula: Wherein, M is the annotation mask, and T is the age-matched anatomical template, C M , C T are the annotation coordinates and the centroid coordinates of the template, and D T is the maximum diameter of the template kidney, ω1 = 0.6, ω2 = 0.4; Morphological rationality index M P The calculation formula is as follows: Where, A is the area of the marked area, P is the perimeter of the marked area, and σ r / μ r is the ratio of the standard deviation in the radial direction to the mean, S k is the Hausdorff distance similarity with the standard ellipsoid, and Sigmoid is a mathematical transformation function; Expert consensus degree E C The calculation formula is as follows: where p i is the annotation probability map of the i-th expert, D KL is the Kullback-Leibler divergence, N is the number of pediatric imaging diagnosis experts participating in the annotation and N > 3; Clinical consistency factor C f The calculation formula is as follows: Where TP is the number of true micro-lesions correctly identified in the annotation result; FP is the number of normal tissue regions mislabeled as lesions in the annotation result; FN is the number of micro-lesions that actually exist but are not annotated; and λ1 is the missed diagnosis penalty coefficient.

4. A multi-dimensional detection method for minute lesions in pediatric kidney CT images according to claim 3, characterized in that, The standard comprehensive quality assessment model in Step 2 is: (5) Where Q is the standard comprehensive quality score, and α, β, and γ are all weight coefficients, and the weight coefficients can be dynamically calibrated through the Bayesian optimization algorithm.

5. A multi-dimensional detection method for minute lesions in children's kidney CT images according to claim 4, characterized in that The dynamic calibration model of the weight coefficients is: where θ is the parameter to be optimized, θ = (α, β, γ); θ * is the optimized parameter, Q gold is the gold standard score determined by the expert group, Q is the standard comprehensive quality score, and λ is the adjustment coefficient.

6. A multi-dimensional detection method for minute lesions in pediatric kidney CT images according to claim 1, characterized in that The dynamic age correction model in Step 3 is: where Q final is the standard scoring threshold, and age is the age of the child to be detected.

7. A multi-dimensional detection method for micro-lesions in children's kidney CT images according to claim 1, characterized in that, The formula for quantitatively analyzing the uncertainty in Step 4 is: Where U Q is the quantitative analysis value of uncertainty, ∈ is to prevent zero factors, ∈ = 0.01, μ s is the historical mean of the spatial anatomical fitness, μ M is the historical mean of the morphological rationality index, μ E is the historical mean of the expert consensus degree.

8. A multi-dimensional detection method for minute lesions in pediatric kidney CT images according to claim 7, characterized in that The quality assessment matrix QM in Step 4 is:

9. A multi-dimensional detection method for minute lesions in pediatric kidney CT images according to claim 8, characterized in that, The real-time comprehensive quality assessment model in Step 4 is: Q = ||QM||F·σ max (QM) (10) where ||·|| F is the Frobenius norm, and σ max is the maximum singular value.

Images