Cervical cancer preoperative path planning method and system

Abstract

The invention discloses a cervical cancer preoperative path planning method and system, and the method comprises the steps: determining a tumor region and a vagina guide needle mold of a to-be-treated patient, and determining the preoperative path of the to-be-treated patient according to the position relation between a tumor target region and an endangered organ, the needle passage layout of the vagina guide needle mold, the prescription dose and the endangered organ dose tolerance limit value; a minimum coverage index and a minimum saving index are defined, the minimum coverage index, the minimum saving index, a residence time space gradient index and a needle passage sparse regularization item are taken as optimization targets, a single target function is constructed through a weighting method, and a reverse optimization model is obtained according to the single target function, a needle inserting and implanting position constraint condition and a residence time constraint condition; and an improved particle swarm optimization algorithm is used to carry out iterative solution on the reverse optimization model to obtain a needle insertion planning result, and the improved particle swarm optimization algorithm is obtained by introducing a random greedy zero setting strategy to carry out improvement. The needle inserting and planting planning result comprises the number of inserted and planted needles, the positions of the inserted and planted needles, the residence positions and the residence time.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional implantation and post-installation, and in particular to a preoperative path planning method and system for cervical cancer. Background Art

[0002] Brachytherapy is a method of delivering a miniature radioactive source (such as iridium-192) directly into the patient's tumor for irradiation via a pre-implanted applicator or source delivery catheter using a computer-controlled stepper motor. It has the advantages of high dose near the source and rapid dose drop around the source. While ensuring that the tumor target area receives high-dose irradiation, it can effectively reduce radiation damage to surrounding normal tissues and organs. It is particularly suitable for increased-dose treatment of local tumor lesions such as paracervical invasion.

[0003] However, the existing after-loading image-guided technology can only provide intraoperative lesion localization, lacks preoperative planning of the implantation path, and cannot customize individualized radiotherapy plans according to the size, shape and invasion depth of the tumor before surgery, and determine the number and layout of implantation needles required during surgery based on the dose distribution. In addition, the traditional IPSA and HIPO algorithm models are relatively simple and cannot automatically optimize the number of implantation needles. There are problems such as unreasonable optimization results and too long dwell time for needle channel edge site allocation. The existing response method uses time constraints (dwell time deviation constraint, DTDC) to limit the difference in dwell time of adjacent sites on a single needle channel, ignoring the mutual influence between needle channels; and because the two algorithms were developed at an early time, they are both based on CPU calculations, and the optimization is relatively time-consuming. Summary of the invention

[0004] In order to solve the above technical problems, the embodiments of the present invention provide a method and system for preoperative path planning for cervical cancer, which realizes the scientific planning of the preoperative implantation path for cervical cancer and improves the accuracy of the planning results.

[0005] A first aspect of an embodiment of the present invention provides a preoperative path planning method for cervical cancer, the method comprising:

[0006] Determining a tumor area of ​​a patient to be treated and a vaginal guide needle mold, wherein a plurality of insertion needle holes are arranged in a preset manner in the vaginal guide needle mold;

[0007] According to the positional relationship between the tumor target area and the organ at risk, the needle track layout of the vaginal guide needle mold, the prescription dose and the dose tolerance limit of the organ at risk, the minimum coverage index and the minimum conservation index are defined, wherein the minimum coverage index is the difference between the radiation dose distribution of the tumor area of ​​the patient to be treated and the tumor area, and the minimum conservation index is the difference between the radiation dose of the tumor area of ​​the patient to be treated and the preset tumor radiation dose;

[0008] Taking the minimum coverage index, the minimum saving index, the residence time spatial gradient index and the needle channel sparse regularization term as optimization objectives, a single objective function is constructed by a weighted method, and a reverse optimization model is obtained according to the single objective function, the implantation needle position constraint condition and the residence time constraint condition, wherein the implantation needle position constraint condition is limited by the implantation needle hole in the vaginal guide needle mold;

[0009] The inverse optimization model is iteratively solved using an improved particle swarm optimization algorithm to obtain an implant needle planning result, wherein the improved particle swarm optimization algorithm is improved by introducing a random greedy zeroing strategy, and the implant needle planning result includes the number of implant needles, implant needle positions, residence positions and residence times.

[0010] In a possible implementation manner of the first aspect, the expression of the residence time spatial gradient index is:

[0011]

[0012] Where S is the set of all needle channels of the vaginal guide needle mold, and vector T i is the total dwell time on the i-th needle track, the vector is the time gradient of the adjacent dwelling positions on the i-th needle track, and the vector represents all the dwell times on the nearest needle track of the i-th needle track;

[0013] The expression of the needle channel sparse regularization term is:

[0014] CSR=‖Te‖1

[0015] Where T is the dwell time matrix, is a vector of all ones, ‖·‖1 is the optimal convex approximation of the L1 norm and also the L0 norm ‖·‖0.

[0016] In a possible implementation of the first aspect, a minimum coverage index, a minimum saving index, a residence time spatial gradient index, and a needle channel sparse regularization term are used as optimization objectives, a single objective function is constructed by a weighted method, and an inverse optimization model is obtained according to the single objective function, the needle position constraint condition, and the residence time constraint condition, including:

[0017] Taking the minimum coverage index, minimum conservation index, residence time spatial gradient index and needle channel sparse regularization term as optimization objectives, the multi-objective optimization problem is converted into a single-objective optimization problem through the weighted method to obtain a single-objective function.

[0018] The inverse optimization model is obtained according to the single objective function, the needle position constraint and the dwell time constraint, where the inverse optimization model is:

[0019] maxλ1LCL+λ2LSL-λ3DTSG-λ4CSR

[0020] st:

[0021]

[0022] T∈{t ij |0≤t ij ≤t max ,i=1,2,...,n,j=1,2,...,m}

[0023] T0∈{t i |0≤t i ≤t max , i = 1, 2, ..., l}

[0024] Where LCL is the minimum coverage index of radiation dose, LSL is the minimum saving index of radiation dose, DTSG is the residence time spatial gradient, CSR is the needle track sparse regularization term, λ1, λ2, λ3 and λ4 are non-negative weight parameters, P is the position matrix of the needle track to be planned, is the set of all needle paths to be planned in the vaginal guide needle mold, T is the residence time matrix corresponding to the position matrix of the needle paths to be planned, t max is the maximum residence time, and T0 is the residence time vector of the uterine cavity tube in the vaginal guide needle mold.

[0025] In a possible implementation of the first aspect, the inverse optimization model is iteratively solved using an improved particle swarm optimization algorithm to obtain an implant needle planning result, including:

[0026] Initialize particle population size, particle position, particle velocity, calculate individual optimal position and evaluate individual optimal fitness;

[0027] The fitness value of the particle position is calculated according to the objective function of the inverse optimization model to obtain the fitness value. If the fitness value is greater than the initial optimal fitness value, the particle position is determined to be the latest individual optimal position, and the optimal needle track position at the current iteration is obtained.

[0028] The inertia weight is updated to obtain an updated inertia weight, and the probability weight of the dwelling position on the optimal needle track position at the current iteration is updated using a random greedy zeroing strategy to obtain an updated probability weight, and the optimal dwelling time matrix in the current optimal needle track position according to the updated probability weight is cleared to obtain the optimal needle track position after clearing;

[0029] Using the updated inertia weight and the optimal needle channel position after being cleared, return to the fitness value of the particle position calculated according to the objective function of the inverse optimization model to obtain a new fitness value. If the new fitness value is greater than the fitness value, the new particle position corresponding to the new fitness value is determined as the implant needle planning result. If it is less than the fitness value, the inertia weight and the probability weight of the residence position on the optimal needle channel position are updated. After the residence position time of the new particle position is cleared, continue to return to the fitness value of the particle position calculated according to the objective function of the inverse optimization model until the implant needle planning result is obtained.

[0030] In a possible implementation manner of the first aspect, the probability weight of the dwell position on the optimal needle track position in the current iteration is updated by using the random greedy zeroing strategy to obtain the updated probability weight, including:

[0031] The Gaussian kernel function is used to calculate the similarity matrix between the optimal needle track position in the current iteration and the optimal needle track position in the previous iteration, wherein the calculation formula of the similarity matrix is:

[0032]

[0033] In the formula, Ψ ij is the similarity matrix, p i is the optimal needle track position at the current iteration, p′ j is the optimal needle track position in the last iteration, σ is the bandwidth of the Gaussian kernel function;

[0034] The similarity is obtained according to the similarity matrix. If the similarity is greater than a preset threshold, the probability weight of the dwell position on the current optimal needle track position is updated to obtain an updated probability weight. If the similarity is less than a preset threshold, the preset initial probability weight is used, wherein the updating formula of the probability weight is:

[0035]

[0036] In the formula, Ψ il is the similarity matrix between the optimal needle track position at the lth iteration and the optimal needle track position at the ith iteration, Φ k-1 is the probability weight matrix of the resident position in the k-1th iteration, Φ 0,ij is the preset initial probability weight.

[0037] A second aspect of an embodiment of the present invention provides a preoperative path planning system for cervical cancer, the system comprising:

[0038] A determination module, used to determine a tumor area of ​​a patient to be treated and a vaginal guide needle mold, wherein a plurality of insertion needle holes are distributed in the vaginal guide needle mold in a preset arrangement;

[0039] The first construction module is used to define a minimum coverage index and a minimum conservation index according to the positional relationship between the tumor target area and the organ at risk, the needle track layout of the vaginal guide needle mold, the prescription dose, and the dose tolerance limit of the organ at risk, wherein the minimum coverage index is the difference between the radiation dose distribution of the tumor area of ​​the patient to be treated and the tumor area, and the minimum conservation index is the difference between the radiation dose of the tumor area of ​​the patient to be treated and the preset tumor radiation dose;

[0040] The second construction module is used to construct a single objective function by weighted method with the minimum coverage index, the minimum saving index, the residence time spatial gradient index and the needle channel sparse regularization term as optimization objectives, and obtain a reverse optimization model according to the single objective function, the implantation needle position constraint condition and the residence time constraint condition, wherein the implantation needle position constraint condition is limited by the implantation needle hole in the vaginal guide needle mold;

[0041] The solution module is used to iteratively solve the inverse optimization model using the improved particle swarm optimization algorithm to obtain the implant needle planning result, wherein the improved particle swarm optimization algorithm is improved by introducing a random greedy zeroing strategy, and the implant needle planning result includes the number of implant needles, implant needle positions, residence positions and residence time.

[0042] In a possible implementation manner of the second aspect, the expression of the residence time spatial gradient index is:

[0043]

[0044] Where S is the set of all needle paths to be planned, and vector T i is the total dwell time on the i-th needle track, the vector is the time gradient of the adjacent dwelling positions on the i-th needle track, and the vector represents all the dwell times on the nearest needle track of the i-th needle track;

[0045] The expression of the needle channel sparse regularization term is:

[0046] CSR=‖Te‖1

[0047] Where T is the dwell time matrix, is a vector of all ones, ‖·‖1 is the optimal convex approximation of the L1 norm and also the L0 norm ‖·‖0.

[0048] In a possible implementation of the second aspect, a minimum coverage index, a minimum saving index, a residence time spatial gradient index, and a needle channel sparse regularization term are used as optimization objectives, a single objective function is constructed by a weighted method, and an inverse optimization model is obtained according to the single objective function, an implant needle position constraint, and a residence time constraint, including:

[0049] Taking the minimum coverage index, minimum conservation index, residence time spatial gradient index and needle channel sparse regularization term as optimization objectives, the multi-objective optimization problem is converted into a single-objective optimization problem through the weighted method to obtain a single-objective function.

[0050] The inverse optimization model is obtained according to the single objective function, the needle position constraint and the dwell time constraint, where the inverse optimization model is:

[0051] maxλ1LCL+λ2LSL-λ3DTSG-λ4CSR

[0052] st:

[0053]

[0054] T∈{t ij |0≤t ij ≤t max ,i=1,2,...,n,j=1,2,...,m}

[0055] T0∈{t i |0≤t i ≤t max , i = 1, 2, ..., l}

[0056] Where LCL is the minimum coverage index of radiation dose, LSL is the minimum saving index of radiation dose, DTSG is the residence time spatial gradient, CSR is the needle track sparse regularization term, λ1, λ2, λ3 and λ4 are non-negative weight parameters, P is the position matrix of the needle track to be planned, is the set of all needle paths to be planned in the vaginal guide needle mold, T is the residence time matrix corresponding to the position matrix of the needle paths to be planned, t max is the maximum residence time, and T0 is the residence time vector of the uterine cavity tube in the vaginal guide needle mold.

[0057] A third aspect of an embodiment of the present invention provides a computer device, including:

[0058] Memory for storing computer programs;

[0059] A processor is used to implement the cervical cancer preoperative path planning method as described in the first aspect when executing the computer program.

[0060] A fourth aspect of an embodiment of the present invention provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the preoperative path planning method for cervical cancer according to the first aspect are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1A planning method flow chart of an embodiment of a preoperative path planning method for cervical cancer provided by the present invention;

[0062] Figure 2 A system block diagram of another embodiment of the method for preoperative path planning for cervical cancer provided by the present invention. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] A flow chart of an embodiment of a method for preoperative path planning for cervical cancer provided by an embodiment of the present invention is as follows: Figure 1 , including steps S101 to S104, each step is as follows:

[0065] S101, determining a tumor area of ​​a patient to be treated and a vaginal guide needle mold, wherein a plurality of insertion needle holes are distributed in the vaginal guide needle mold in a preset arrangement.

[0066] In this embodiment, in view of the problems that the traditional IPSA and HIPO algorithm models are relatively simple, cannot automatically optimize the number of implant needles, and the optimization results are not reasonable enough, and the needle channel edge site allocation residence time is too long, a new reverse optimization model is constructed based on the improved vaginal mold and the new generation of navigation-assisted stent for preoperative path planning.

[0067] First, based on the automatic segmentation of cervical cancer tumor target area and organs at risk on high-resolution CT and MRI images, the positional relationship between the target area and organs at risk is obtained, and the tumor area and vaginal guide needle mold of the patient to be treated are determined, wherein multiple insertion needle holes are distributed in the vaginal guide needle mold according to a preset arrangement.

[0068] Then, combined with the traditional target dose coverage indicators and the dose limits of organs at risk, that is, the prescription dose and the normal tissue dose tolerance limit, preparation is made for the next step of determining the indicators to be optimized.

[0069] It should be noted that the mold body in the vaginal guide needle mold is a mushroom-shaped column with 38 optional holes. The column is 102mm long and the insertable part is 90mm long. The mushroom head has a long diameter of 40mm and a short diameter of 20mm. After deducting the edge of the template, the effective radius is about 17.5mm. The center is adapted to the uterine cavity tube, and there are 12 parallel needle channels C1-C arranged in a dial-like manner around the uterine cavity tube. 12 The expansion parts on both sides are equipped with parallel needle channels A1-A that are symmetrical on both sides. 17and oblique needle channels B1-B8, totaling 38 needle channels.

[0070] S102. Define a minimum coverage index and a minimum conservation index based on the positional relationship between the tumor target area and the organ at risk, the needle track layout of the vaginal guide needle mold, the prescription dose, and the dose tolerance limit of the organ at risk, wherein the minimum coverage index is the difference between the radiation dose distribution in the tumor area of ​​the patient to be treated and the tumor area, and the minimum conservation index is the difference between the radiation dose in the tumor area of ​​the patient to be treated and the preset tumor radiation dose.

[0071] In this embodiment, after the tumor area is determined using medical imaging technology, the size, shape, location of the tumor and its relationship with surrounding tissues can be clearly displayed, and then a radiotherapy plan is formulated based on the location of the tumor area and the organ at risk. The minimum coverage index and the minimum conservation index are first defined, wherein the minimum coverage index is the difference between the radiation dose distribution of the tumor area of ​​the patient to be treated and the tumor area, and the minimum conservation index is the difference between the radiation dose of the tumor area of ​​the patient to be treated and the preset tumor radiation dose.

[0072] S103. Taking the minimum coverage index, the minimum saving index, the residence time spatial gradient index and the needle channel sparse regularization term as optimization objectives, a single objective function is constructed by a weighted method, and a reverse optimization model is obtained according to the single objective function, the insertion needle position constraint and the residence time constraint, wherein the insertion needle position constraint is limited by the insertion needle hole in the vaginal guide needle mold.

[0073] In some embodiments, based on the target area and risk organ dose constraints, the needle track sparsity regularization option and the concept of dwell time spatial gradient optimization are introduced to achieve automatic optimization of the number, position and dwell time of needle tracks. The specific process is as follows:

[0074] Assuming that at most n needle insertion holes need to be selected outside the uterine cavity tube, for the i-th needle channel hole of the vaginal guide needle mold, let its hole coordinates be (x i ,y i ,z i ), then the insertion needle track position and its corresponding needle track direction can be defined as:

[0075] p i =(x i ,y i ,z i ),v i =(α i ,β i ,γ i ),i=1,2,...,n.

[0076] Assume the needle path length is L, and the radiation source moves forward along the catheter in increments of Δc each time. Then in the i-th implant needle, there will be at most The mapping relationship between the coordinates of the jth dwelling position on the i-th needle channel and the needle channel hole position is:

[0077] h ij =p i +jΔcv i ,i=1,2,...,n,j=1,2,...,m.

[0078] Where m is the number of dwell positions, Δc is the increment of the radiation source, and p i This is the location for the needle insertion.

[0079] In addition, the needle path position matrix outside the uterine cavity is recorded The corresponding residence time matrix is In particular, since the position of the uterine cavity tube is fixed, the position of the needle track is considered to be known and there are l available dwell positions, then the dwell time vector of the uterine cavity tube can be defined as

[0080] The “residence time spatial gradient index and needle track sparse regularization term” in step S103 includes but is not limited to the following steps:

[0081] The expression of the residence time spatial gradient index is:

[0082]

[0083] Where S is the set of all needle paths to be planned, and vector T i is the total dwell time on the i-th needle track, the vector is the time gradient of the adjacent dwelling positions on the i-th needle track, and the vector represents all the dwell times on the nearest needle track of the i-th needle track;

[0084] The expression of the needle channel sparse regularization term is:

[0085] CSR=‖Te‖1

[0086] Where T is the dwell time matrix, is a vector of all ones, ‖·‖1 is the L1 norm and also the best convex approximation of the L0 norm ‖·‖0.

[0087] In this embodiment, the dwell time spatial gradient index comprehensively evaluates the time uniformity of multiple needle channels, and optimizes the index to improve the uniformity of the overall dwell time distribution during multi-needle channel optimization. The expression of the dwell time spatial gradient is:

[0088]

[0089] Where S is the set of all the needle paths to be planned, and the vector T i is the total dwell time on the i-th needle track, the vector is the time gradient of the adjacent dwelling positions on the i-th needle track, and the vector represents all the dwell times on the nearest needle tracks of the i-th needle track.

[0090] In order to reduce the redundancy of needle channels and improve the rationality of needle placement to alleviate the pain of patients, needle channel sparse regularization is introduced in the inverse optimization model. According to the definition, the number of needle channels can be recorded as: ‖Te‖0, where T is the residence time matrix, is a vector of all 1s, and ‖·‖0 is the L0 norm. However, due to its non-convexity, the L0 norm will bring certain difficulties to the optimization solution, so its optimal convex approximation L1 norm is used to replace the L0 norm, which is called sparse regularization. Finally, the following CSR index is added to the objective function of the optimization model to achieve reasonable optimization of the number of needle channels:

[0091]

[0092] Where S is the set of all the needle paths to be planned, and the vector T i is the total dwell time on the i-th needle track, the vector is the time gradient of the adjacent dwelling positions on the i-th needle track, and the vector represents all the dwell times on the nearest needle tracks of the i-th needle track.

[0093] Intracavitary combined with interstitial implantation for locally advanced cervical cancer, the overall dose distribution of afterloading is defined as:

[0094] D(P,T,T0)=D1(P,T)+D2(T0),

[0095] Among them, D(P,T,T0) represents the overall dose distribution, D1(P,T) is the parauterine insertion dose, which is determined by the parauterine insertion needle position P and its residence time T, and D2(T0) is the uterine cavity tube dose, which is determined by the uterine cavity tube residence time T0.

[0096] In some embodiments, in step S103, "taking the minimum coverage index, the minimum saving index, the residence time spatial gradient index and the needle channel sparse regularization term as optimization objectives, constructing a single objective function by a weighted method, and obtaining a reverse optimization model according to the single objective function, the implant needle position constraint and the residence time constraint" includes but is not limited to the following steps:

[0097] Taking the minimum coverage index, minimum conservation index, residence time spatial gradient index and needle channel sparse regularization term as optimization objectives, the multi-objective optimization problem is converted into a single-objective optimization problem through the weighted method to obtain a single-objective function.

[0098] The inverse optimization model is obtained according to the single objective function, the needle position constraint and the dwell time constraint, where the inverse optimization model is:

[0099] maxλ1LCL+λ2LSL-λ3DTSG-λ4CSR

[0100] st:

[0101]

[0102] T∈{t ij |0≤t ij ≤t max ,i=1,2,...,n,j=1,2,...,m}

[0103] T0∈{t i |0≤t i ≤t max , i = 1, 2, ..., l}

[0104] Where LCL is the minimum coverage index of radiation dose, LSL is the minimum saving index of radiation dose, DTSG is the residence time spatial gradient, CSR is the needle track sparse regularization term, λ1, λ2, λ3 and λ4 are non-negative weight parameters, P is the position matrix of the needle track to be planned, is the set of all needle paths to be planned in the vaginal guide needle mold, T is the residence time matrix corresponding to the position matrix of the needle paths to be planned, t max is the maximum residence time, and T0 is the residence time vector of the uterine cavity tube in the vaginal guide needle mold.

[0105] In this embodiment, a new inverse optimization model is constructed based on the designed vaginal mold and the new generation of navigation-assisted stent, with the minimum radiation dose coverage index, the minimum radiation dose saving index, the residence time spatial gradient and the needle channel sparse regularization as the optimization targets. The target area coverage index is used to prevent too few implanted needles, and the needle channel sparse regularization is used to avoid too many needles. The two are played together, so that the final treatment plan can achieve a better balance between the dosimetric index and the number of needle channels.

[0106] The complex multi-objective optimization is transformed into a single-objective optimization problem through the weighted method, and the inverse optimization model is constructed by combining the position and residence time constraints of the needle track to be planned. The inverse optimization model is:

[0107] maxλ1LCL+λ2LSL-λ3DTSG-λ4CSR

[0108] st:

[0109]

[0110] T∈{t ij |0≤t ij ≤t max ,i=1,2,...,n,j=1,2,...,m}

[0111] T0∈{t i |0≤t i ≤t max , i = 1, 2, ..., l}

[0112] Wherein, LCL is the minimum coverage index of radiation dose, LSL is the minimum saving index of radiation dose, DTSG is the residence time spatial gradient, CSR is the needle track sparse regularization, λ1, λ2, λ3 and λ4 are non-negative weight parameters, P is the position matrix of the needle track to be planned, is the set of all the needle paths to be planned in the vaginal guide needle mold, T is the residence time matrix corresponding to the position matrix of the needle paths to be planned, t max is the maximum residence time, and T0 is the residence time vector of the uterine cavity tube in the vaginal guide needle mold.

[0113] The constraints of the inverse optimization model include the position constraint of the implant needle, the residence time constraint and the radiation dose constraint. The radiation dose constraint is the overall dose distribution of afterloading, including the parauterine implant dose and the uterine cavity dose.

[0114] S104. Use the improved particle swarm optimization algorithm to iteratively solve the inverse optimization model to obtain the implant needle planning result, wherein the improved particle swarm optimization algorithm is improved by introducing a random greedy zeroing strategy, and the implant needle planning result includes the number of implant needles, implant needle positions, residence positions and residence time.

[0115] In this embodiment, a complex multi-objective optimization model is established to comprehensively optimize multiple indicators. In addition, in order to achieve sparse optimization of the needle channel to reduce the number of needles, needle channel sparsity regularization is also used in the objective function of the optimization model. Due to the complexity and non-smoothness of the objective function, the traditional gradient-based optimization method is relatively expensive to calculate. In addition, the optimization model with sparse regularization terms usually lacks effective gradient information at the optimal solution, or is prone to large numerical errors when estimating the gradient, resulting in unstable convergence, falling into local optimal solutions, and skipping sparse optimal solutions. Therefore, by jumping out of the traditional algorithm framework, the particle swarm optimization algorithm is adopted, combined with GPU parallel computing to solve the established optimization model more efficiently and accurately. Furthermore, in order to improve the global search performance of the particle swarm optimization algorithm and the ability to solve sparse optimal solutions, the particle swarm optimization algorithm is improved by setting an inertia weight update strategy with oscillation attenuation and a random greedy zeroing strategy.

[0116] In some embodiments, the step S104 of "using the improved particle swarm optimization algorithm to iteratively solve the inverse optimization model to obtain the implant needle planning result" includes but is not limited to the following steps:

[0117] Initialize particle population size, particle position, particle velocity, calculate individual optimal position and evaluate individual optimal fitness;

[0118] The fitness value of the particle position is calculated according to the objective function of the inverse optimization model to obtain the fitness value. If the fitness value is greater than the initial optimal fitness value, the particle position is determined to be the latest individual optimal position, and the optimal needle track position at the current iteration is obtained.

[0119] The inertia weight is updated to obtain an updated inertia weight, and the probability weight of the dwelling position on the optimal needle track position at the current iteration is updated using a random greedy zeroing strategy to obtain an updated probability weight, and the optimal dwelling time matrix in the current optimal needle track position according to the updated probability weight is cleared to obtain the optimal needle track position after clearing;

[0120] Using the updated inertia weight and the optimal needle channel position after being cleared, return to the fitness value of the particle position calculated according to the objective function of the inverse optimization model to obtain a new fitness value. If the new fitness value is greater than the fitness value, the new particle position corresponding to the new fitness value is determined as the implant needle planning result. If it is less than the fitness value, the inertia weight and the probability weight of the residence position on the optimal needle channel position are updated. After the residence position time of the new particle position is cleared, continue to return to the fitness value of the particle position calculated according to the objective function of the inverse optimization model until the implant needle planning result is obtained.

[0121] In this embodiment, by comparing the particle swarm optimization algorithm using the random greedy zeroing strategy with the original particle swarm optimization algorithm, and analyzing and verifying the feasibility and advancement of the improvement from aspects such as convergence speed, sparsity of the solution, and optimality of the solution, there are obvious improvements. The steps of the improved particle swarm optimization algorithm are as follows:

[0122] Input: particle population size K, particle dimension D, maximum number of iterations i M , inertia weight ω pso , learning factors c1, c2. Output: population optimal solution p gb .

[0123] As shown in Table 1, first, the particle position is initialized within the allowable range. Initialize particle speed within the allowed range Initialize the optimal position of the individual is a zero vector, initializing the individual's optimal fitness is +∞. Initialize the optimal position p of the population gb is a zero vector, and the optimal fitness of the initialized population is f gb .

[0124] Table 1 Improved particle swarm optimization algorithm steps

[0125]

[0126]

[0127] Calculate the current particle position x according to the objective function of the optimization model k The fitness f, if Otherwise Then update f gb and p gb , update the inertia weight ω according to the oscillation decay strategy pso , and then execute the random greedy zeroing strategy.

[0128] In some embodiments, step S104 of "updating the probability weight of the dwell position on the optimal needle track position in the current iteration by using a random greedy zeroing strategy to obtain an updated probability weight" includes but is not limited to the following steps:

[0129] The Gaussian kernel function is used to calculate the similarity matrix between the optimal needle track position in the current iteration and the optimal needle track position in the previous iteration, where the calculation formula of the similarity matrix is:

[0130]

[0131] In the formula, Ψ ij is the similarity matrix, p iis the current optimal needle track position, p j ′ is the optimal needle track position in the last iteration, σ is the bandwidth of the Gaussian kernel function;

[0132] The similarity is obtained according to the similarity matrix. If the similarity is greater than the preset threshold, the probability weight of the dwelling position on the current optimal needle track position is updated to obtain the updated probability weight. If the similarity is less than the preset threshold, the preset initial probability weight is used, where the probability weight update formula is:

[0133]

[0134] In the formula, Ψ il is the similarity matrix between the optimal needle track position at the lth iteration and the optimal needle track position at the ith iteration, Φ k-1 is the probability weight before updating, Φ 0,ij is the preset initial probability weight.

[0135] In this embodiment, since the inverse optimization model contains sparse regularization terms, the speed at which the traditional particle swarm optimization algorithm converges to the sparse optimal solution on this type of sparse optimization problem is usually slow. In order to improve the speed at which the particle swarm optimization algorithm obtains the sparse optimal solution, a random greedy zeroing strategy based on historical similarity is designed by drawing on the attention mechanism in machine learning, which is used to sparse the population optimal solution of the particle swarm, thereby guiding the particle swarm to search for the sparse optimal solution faster. The core idea of ​​the random greedy zeroing strategy is: a probability weight is assigned to each residence position on each needle path, and the probability is updated according to the change of the optimal needle path in each iteration of the particle swarm optimization. According to the size of the probability weight, a residence position is randomly selected in each current optimal needle path to reset its residence time. If a better solution is generated after random zeroing, the new solution is used as the new particle swarm optimal solution, otherwise the weight of the corresponding position is reset.

[0136] Define the probability weight matrix of the residence position in the kth iteration and its default probability weight matrix Usually the default probability weight matrix can be selected as The specific implementation steps of the random greedy zeroing strategy are as follows:

[0137] Get the current particle swarm optimal needle track position P gb =[p1,p2,…,p n ] and the optimal needle track position P′ of the particle swarm in the previous iteration gb =[p′1,p′2,…,p′ n ], using Gaussian kernel function to calculate the similarity matrix The calculation formula of the similarity matrix is:

[0138]

[0139] In the formula, Ψ ij is the similarity matrix, p i is the current optimal needle track position, p′ j is the optimal needle track position in the previous iteration, and σ is the bandwidth of the Gaussian kernel function.

[0140] The similarity matrix affects the update strategy of the probability weight matrix. The higher the similarity of the optimal needle paths between the two iterations, the more similar the probability weights used. If the similarity is lower than the threshold ∈, the default probability weights are used. Therefore, the iterative update format of the weight matrix is ​​as follows:

[0141]

[0142] In the formula, Ψ il is the similarity matrix between the optimal needle track position at the lth iteration and the optimal needle track position at the ith iteration, Φ k-1 is the probability weight matrix of the resident position in the k-1th iteration, Φ 0,ij is the preset initial probability weight.

[0143] According to the current probability weight matrix Φ k , randomly set zeros for each row of the current optimal residence time matrix, and evaluate the fitness of the new solution. If the fitness of the new solution is higher, the particle swarm optimal solution is updated, otherwise the probability weight of the corresponding position is cleared.

[0144] In addition, the Gaussian kernel function bandwidth and similarity threshold can be set according to actual needs.

[0145] To verify the effectiveness of the new optimization model and particle swarm optimization algorithm for rapid solution, 100 patients with stage IIB-IVA (FIGO 2018 stage) cervical cancer who were initially diagnosed without distant metastasis and received radical radiotherapy from January 2018 to December 2023 were randomly selected from a certain unit. The medical Oncentra afterloading treatment planning system (v.4.6.3, Elekta) IPSA algorithm was used as a reference. The differences between the two in terms of the number of implants, optimization time, and dose distribution (including the maximum dose, average dose, and dose-volume parameters such as Vx and Dy in the tumor target area and each organ at risk) were statistically analyzed in parallel. The results showed that by constructing a new afterloading reverse optimization model and applying the particle swarm optimization algorithm and GPU acceleration technology to achieve rapid solution, the average plan optimization time was less than 40 seconds, and the complex plan optimization time was less than 60 seconds.

[0146] It should be understood that although Figure 1The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0147] In some embodiments, Figure 2 As shown, it shows a block diagram of a cervical cancer preoperative path planning system 200 provided in an embodiment of the present application, including: a determination module 201, a first construction module 202, a second construction module 203 and a solution module 204, wherein:

[0148] The determination module 201 is used to determine the tumor area of ​​the patient to be treated and the vaginal guide needle mold, wherein a plurality of insertion needle holes are distributed in the vaginal guide needle mold according to a preset arrangement;

[0149] The first construction module 202 is used to define a minimum coverage index and a minimum conservation index according to the positional relationship between the tumor target area and the organ at risk, the needle track layout of the vaginal guide needle mold, the prescription dose, and the dose tolerance limit of the organ at risk, wherein the minimum coverage index is the difference between the radiation dose distribution of the tumor area of ​​the patient to be treated and the tumor area, and the minimum conservation index is the difference between the radiation dose of the tumor area of ​​the patient to be treated and the preset tumor radiation dose;

[0150] The second construction module 203 is used to construct a single objective function by weighting method with the minimum coverage index, the minimum saving index, the residence time spatial gradient index and the needle channel sparse regularization term as optimization targets, and obtain a reverse optimization model according to the single objective function, the implantation needle position constraint condition and the residence time constraint condition, wherein the implantation needle position constraint condition is limited by the implantation needle hole in the vaginal guide needle mold;

[0151] The solution module 204 is used to iteratively solve the inverse optimization model using the improved particle swarm optimization algorithm to obtain the implantation needle planning result, wherein the improved particle swarm optimization algorithm is improved by introducing a random greedy zeroing strategy, and the implantation needle planning result includes the number of implantation needles, the implantation needle position, the residence position and the residence time.

[0152] In some embodiments, the expression of the residence time spatial gradient index is:

[0153]

[0154] Where S is the set of all needle paths to be planned, and vector T i is the total dwell time on the i-th needle track, the vector is the time gradient of the adjacent dwelling positions on the i-th needle track, and the vector represents all the dwell times on the nearest needle track of the i-th needle track;

[0155] The expression of the needle channel sparse regularization term is:

[0156] CSR=‖Te‖1

[0157] Where T is the dwell time matrix, is a vector of all ones, ‖·‖1 is the L1 norm and also the best convex approximation of the L0 norm ‖·‖0.

[0158] In some embodiments, the minimum coverage index, the minimum saving index, the residence time spatial gradient index and the needle channel sparse regularization term are used as optimization targets, a single objective function is constructed by a weighted method, and a reverse optimization model is obtained according to the single objective function, the needle position constraint and the residence time constraint, including:

[0159] Taking the minimum coverage index, minimum conservation index, residence time spatial gradient index and needle channel sparse regularization term as optimization objectives, the multi-objective optimization problem is converted into a single-objective optimization problem through the weighted method to obtain a single-objective function.

[0160] The inverse optimization model is obtained according to the single objective function, the needle position constraint and the dwell time constraint, where the inverse optimization model is:

[0161] maxλ1LCL+λ2LSL-λ3DTSG-λ4CSR

[0162] st:

[0163]

[0164] T∈{t ij |0≤t ij ≤t max ,i=1,2,...,n,j=1,2,...,m}

[0165] T0∈{t i |0≤t i ≤t max , i = 1, 2, ..., l}

[0166] Where LCL is the minimum coverage index of radiation dose, LSL is the minimum saving index of radiation dose, DTSG is the residence time spatial gradient, CSR is the needle track sparse regularization term, λ1, λ2, λ3 and λ4 are non-negative weight parameters, P is the position matrix of the needle track to be planned, is the set of all needle paths to be planned in the vaginal guide needle mold, T is the residence time matrix corresponding to the position matrix of the needle paths to be planned, t max is the maximum residence time, and T0 is the residence time vector of the uterine cavity tube in the vaginal guide needle mold.

[0167] The specific implementation of the preoperative path planning system for cervical cancer is basically the same as the specific implementation of the preoperative path planning method for cervical cancer described above, and will not be repeated here.

[0168] In one embodiment of the present application, a computer device is provided, which includes a memory and a processor, wherein a computer program is stored in the memory, and the above steps are implemented when the processor executes the computer program; the computer device provided in this embodiment has an implementation principle and technical effects similar to those of the above method embodiments, and will not be repeated here.

[0169] In one embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and the above steps are implemented when the computer program is executed by a processor; the computer-readable storage medium provided in this embodiment has an implementation principle and technical effects similar to those of the above method embodiment, and will not be repeated here.

[0170] Compared with the prior art, the present invention has the following beneficial effects:

[0171] The present invention determines the tumor area and vaginal guide needle mold of the patient to be treated, defines the minimum coverage index and the minimum conservation index according to the tumor area, the position of the endangered organ and the treatment plan, takes the minimum coverage index, the minimum conservation index, the residence time spatial gradient index and the needle channel sparse regularization term as the optimization objectives, constructs a single objective function through a weighted method, obtains an inverse optimization model according to the single objective function, the implant needle position constraint condition and the residence time constraint condition, wherein the implant needle position constraint condition is limited by the implant needle hole in the vaginal guide needle mold; uses an improved particle swarm optimization algorithm to iteratively solve the inverse optimization model to obtain an implant needle planning result, wherein the improved particle swarm optimization algorithm is improved by introducing a random greedy zeroing strategy, and the implant needle planning result includes the number of implant needles, the implant needle position, the residence position and the residence time.

[0172] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0173] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A preoperative path planning method for cervical cancer, characterized in that: include: Determining a tumor area of ​​a patient to be treated and a vaginal guide needle mold, wherein a plurality of insertion needle holes are arranged in a preset manner in the vaginal guide needle mold; According to the positional relationship between the tumor target area and the organ at risk, the needle track layout of the vaginal guide needle mold, the prescription dose and the dose tolerance limit of the organ at risk, the minimum coverage index and the minimum conservation index are defined, wherein the minimum coverage index is the difference between the radiation dose distribution of the tumor area of ​​the patient to be treated and the tumor area, and the minimum conservation index is the difference between the radiation dose of the tumor area of ​​the patient to be treated and the preset tumor radiation dose; Taking the minimum coverage index, the minimum saving index, the residence time spatial gradient index and the needle channel sparse regularization term as optimization objectives, a single objective function is constructed by a weighted method, and a reverse optimization model is obtained according to the single objective function, the implantation needle position constraint condition and the residence time constraint condition, wherein the implantation needle position constraint condition is limited by the implantation needle hole in the vaginal guide needle mold; The inverse optimization model is iteratively solved using an improved particle swarm optimization algorithm to obtain an implant needle planning result, wherein the improved particle swarm optimization algorithm is improved by introducing a random greedy zeroing strategy, and the implant needle planning result includes the number of implant needles, implant needle positions, residence positions and residence times.

2. The method for preoperative path planning for cervical cancer according to claim 1, characterized in that: The expression of the residence time spatial gradient index is: Where S is the set of all needle channels of the vaginal guide needle mold, and the vector T i is the total dwell time on the i-th needle track, the vector is the time gradient of the adjacent dwelling positions on the i-th needle track, and the vector represents all the dwell times on the nearest needle track of the i-th needle track; The expression of the needle channel sparse regularization term is: CSR=‖Te‖1 Where T is the dwell time matrix, is a vector of all ones, ‖·‖1 is the optimal convex approximation of the L1 norm and also the L0 norm ‖·‖0.

3. The method for preoperative path planning for cervical cancer according to claim 1, characterized in that: The minimum coverage index, the minimum saving index, the residence time spatial gradient index and the needle channel sparse regularization term are used as optimization targets, a single objective function is constructed by a weighted method, and a reverse optimization model is obtained according to the single objective function, the implantation needle position constraint condition and the residence time constraint condition, including: Taking the minimum coverage index, minimum conservation index, residence time spatial gradient index and needle channel sparse regularization term as optimization objectives, the multi-objective optimization problem is converted into a single-objective optimization problem through the weighted method to obtain a single-objective function. The reverse optimization model is obtained according to the single objective function, the needle position constraint and the residence time constraint, wherein the reverse optimization model is: maxλ1LCL+λ2LSL-λ3DTSG-λ4CSR st: T∈{t ij |0≤t ij ≤t max ,i=1,2,...,n,j=1,2,...,m} T0∈{t i |0≤t i ≤t max ,i=1,2,...,l} Wherein, LCL is the minimum coverage index of radiation dose, LSL is the minimum saving index of radiation dose, DTSG is the residence time spatial gradient, CSR is the needle track sparse regularization term, λ1, λ2, λ3 and λ4 are non-negative weight parameters, P is the position matrix of the needle track to be planned, is the set of all the needle paths to be planned in the vaginal guide needle mold, T is the residence time matrix corresponding to the position matrix of the needle paths to be planned, t max is the maximum residence time, and T0 is the residence time vector of the uterine cavity tube in the vaginal guide needle mold.

4. The method for preoperative path planning for cervical cancer according to claim 1, characterized in that: The improved particle swarm optimization algorithm is used to iteratively solve the inverse optimization model to obtain the implant needle planning result, including: Initialize particle population size, particle position, particle velocity, calculate individual optimal position and evaluate individual optimal fitness; Calculating the fitness value of the particle position according to the objective function of the inverse optimization model to obtain a fitness value, and if the fitness value is greater than the initial optimal fitness value, determining the particle position as the latest individual optimal position, and obtaining the optimal needle track position at the current iteration; The inertia weight is updated to obtain an updated inertia weight, the probability weight of the dwell position on the optimal needle track position at the current iteration is updated using the random greedy zeroing strategy to obtain an updated probability weight, and the optimal dwell time matrix in the current optimal needle track position is cleared according to the updated probability weight to obtain the cleared optimal needle track position; Utilizing the updated inertia weight and the optimal needle channel position after being cleared, return to the fitness value of the particle position calculated according to the objective function of the inverse optimization model to obtain a new fitness value. If the new fitness value is greater than the fitness value, determine that the new particle position corresponding to the new fitness value is the implantation needle planning result. If it is less than the fitness value, update the inertia weight and the probability weight of the residence position on the optimal needle channel position, clear the residence position time of the new particle position, and continue to return to the fitness value of the particle position calculated according to the objective function of the inverse optimization model until the implantation needle planning result is obtained.

5. The method for preoperative path planning for cervical cancer according to claim 4, characterized in that: The method of updating the probability weight of the dwell position on the optimal needle track position in the current iteration by using the random greedy zeroing strategy to obtain the updated probability weight includes: The Gaussian kernel function is used to calculate the similarity matrix between the optimal needle track position in the current iteration and the optimal needle track position in the previous iteration, wherein the calculation formula of the similarity matrix is: In the formula, Ψ ij is the similarity matrix, p i is the optimal needle track position at the current iteration, p j ′ is the optimal needle track position in the last iteration, σ is the bandwidth of the Gaussian kernel function; The similarity is obtained according to the similarity matrix. If the similarity is greater than a preset threshold, the probability weight of the dwell position on the current optimal needle track position is updated to obtain an updated probability weight. If the similarity is less than a preset threshold, the preset initial probability weight is used, wherein the updating formula of the probability weight is: In the formula, Ψ il is the similarity matrix between the optimal needle track position at the lth iteration and the optimal needle track position at the ith iteration, Φ k-1 is the probability weight matrix of the resident position in the k-1th iteration, Φ 0,ij is the preset initial probability weight.

6. A cervical cancer preoperative path planning system, characterized in that: include: A determination module, used to determine the tumor area of ​​the patient to be treated and the vaginal guide needle mold, wherein the vaginal guide needle mold has a plurality of insertion needle holes distributed in a preset arrangement; The first construction module is used to define a minimum coverage index and a minimum conservation index according to the positional relationship between the tumor target area and the organ at risk, the needle track layout of the vaginal guide needle mold, the prescription dose, and the dose tolerance limit of the organ at risk, wherein the minimum coverage index is the difference between the radiation dose distribution of the tumor area of ​​the patient to be treated and the tumor area, and the minimum conservation index is the difference between the radiation dose of the tumor area of ​​the patient to be treated and the preset tumor radiation dose; The second construction module is used to construct a single objective function by weighted method with the minimum coverage index, the minimum saving index, the residence time spatial gradient index and the needle channel sparse regularization term as optimization objectives, and obtain a reverse optimization model according to the single objective function, the implantation needle position constraint condition and the residence time constraint condition, wherein the implantation needle position constraint condition is limited by the implantation needle hole in the vaginal guide needle mold; A solution module is used to iteratively solve the inverse optimization model using an improved particle swarm optimization algorithm to obtain an implant needle planning result, wherein the improved particle swarm optimization algorithm is improved by introducing a random greedy zeroing strategy, and the implant needle planning result includes the number of implant needles, implant needle positions, residence positions and residence times.

7. The cervical cancer preoperative path planning system according to claim 6, characterized in that: The expression of the residence time spatial gradient index is: Where S is the set of all the needle paths to be planned, and the vector T i is the total dwell time on the i-th needle track, the vector is the time gradient of the adjacent dwelling positions on the i-th needle track, and the vector represents all the dwell times on the nearest needle track of the i-th needle track; The expression of the needle channel sparse regularization term is: CSR=‖Te‖1 Where T is the dwell time matrix, is a vector of all ones, ‖·‖1 is the optimal convex approximation of the L1 norm and also the L0 norm ‖·‖0.

8. The cervical cancer preoperative path planning system according to claim 6, characterized in that: The minimum coverage index, the minimum saving index, the residence time spatial gradient index and the needle channel sparse regularization term are used as optimization targets, a single objective function is constructed by a weighted method, and a reverse optimization model is obtained according to the single objective function, the implantation needle position constraint condition and the residence time constraint condition, including: Taking the minimum coverage index, minimum conservation index, residence time spatial gradient index and needle channel sparse regularization term as optimization objectives, the multi-objective optimization problem is converted into a single-objective optimization problem through the weighted method to obtain a single-objective function. The reverse optimization model is obtained according to the single objective function, the needle position constraint and the residence time constraint, wherein the reverse optimization model is: maxλ1LCL+λ2LSL-λ3DTSG-λ4CSR st: T∈{t ij |0≤t ij ≤t max ,i=1,2,...,n,j=1,2,...,m} T0∈{t i |0≤t i ≤t max ,i=1,2,...,l} Wherein, LCL is the minimum coverage index of radiation dose, LSL is the minimum saving index of radiation dose, DTSG is the residence time spatial gradient, CSR is the needle track sparse regularization term, λ1, λ2, λ3 and λ4 are non-negative weight parameters, P is the position matrix of the needle track to be planned, is the set of all the needle paths to be planned in the vaginal guide needle mold, T is the residence time matrix corresponding to the position matrix of the needle paths to be planned, t max is the maximum residence time, and T0 is the residence time vector of the uterine cavity tube in the vaginal guide needle mold.

9. A computer device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the method for preoperative path planning for cervical cancer as claimed in any one of claims 1 to 5 when executing the computer program.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the preoperative path planning method for cervical cancer according to any one of claims 1 to 5 are implemented.

Images