A Study On The Optimization Algorithm Of Aperture Shape Generation For IMRT

Radiotherapy is one of the three conventional treatments for cancer.More than half of cancer patients need to receive radiotherapy in the process of treatment.In the process of radiotherapy,the chromosome of cancer cells is destroyed by the energy of radiation to kill cancer cells and treat cancer.Traditional radiotherapy will inevitably irradiate the normal tissue around the target and cause radiation damage.In order to reduce the dose received by normal tissue in the process of radiotherapy and reduce the normal tissue complication probability(NTCP),intensity-modulated radiation therapy(IMRT)came into being.IMRT uses a variety of intensity modulation techniques,according to the radiation treatment plan obtained by the optimization method to set the shapes and weights of the apertures,so that the high dose area of the radiation on the human body coincides with the target,reducing the dose received by the normal tissue.Based on multi-leaf collimator static intensity modulation technology for step-and-shoot IMRT,this study carried out in-depth research on the related aperture shape generation methods,and the main work is as follows:(1)In the IMRT plan optimization,the objective function of physical criteria can not accurately describe the biological response of the biological tissues under the non-uniform dose distribution,limited memory BFGS(L-BFGS)algorithm can not directly solve the IMRT plan optimization problem.Aiming at those problems,a new IMRT plan optimization method based on biological criteria is proposed,and applies it to solve the fluence map optimization(FMO)problem in the process of step-and-shoot IMRT realized by two-step method.The proposed method first introduces the biological criteria into the construction of the total objective function of the IMRT plan optimization problem,studies the characteristics of the biological criteria and the functional properties of the corresponding subobjective function,and compares the effects of the objective function based on physical criteria and the objective function based on biological criteria on the optimization results.Then,L-BFGS for bound constrained(L-BFGS-B)optimization algorithm is used to solve the FMO problem,and the similarities and differences between L-BFGS algorithm and L-BFGS-B algorithm in the process of solving the FMO problem are compared.The results show that introducing the subobjective function based on biological criteria into the total objective function of the optimization problem can more accurately reflect the radiation dosimetry effect of biological tissues,reasonably predict the NTCP and give feedback;using L-BFGS-B algorithm to directly solve the IMRT plan optimization problem can further reduce the calculation amount of solving the problem.(2)In the process of step-and-shoot IMRT realized by direct aperture optimization(DAO),a network flow is adopted to solve the pricing problem of the generic column generation(CG)approach to obtain a deliverable aperture.However,excessive computation results from the direct use of a network flow,and a decline in plan quality may result from the direct determination of the leaf position using the gradient information.In order to solve those problems,a CG approach based on region growth is proposed.First,the gradients in the aperture gradient map are transformed nonlinearly,the relationship between the beamlet gradient and the aperture shape is studied.Second,the positions of the continuous negative gradient regions in each row of the gradient map are determined,the continuous negative gradient regions in the gradient map are segmented,and the proposed method performance is analyzed.Then,these gradients in the processed gradient map are added to the aperture network,which is solved as a shortest path problem.Finally,the deliverable aperture is obtained and added to the treatment plan,and the weights of the generated apertures are optimized.The research shows that compared with the generic CG,the proposed method can reduce the computation of generating aperture without changing the algorithm structure,and improve the quality of the treatment plan.(3)In generic DAO,the direction of negative gradient descent is generally used to determine the aperture shape.However,this strategy can reduce the convergence rate,especially near the optimal value.Aperture shape generation based on the direction of gradient descent with momentum is proposed to accelerate the convergence process of the generic CG using negative gradient descent direction to search the aperture shape.In the process of aperture shape generation,firstly,the aperture gradient map is obtained,and the principles of acceleration of search process and reduction of search direction oscillation by momentum gradient descent method are studied compared with the steepest descent method.Then,the momentum gradient is calculated according to the gradient in the gradient map,and the characteristics of the momentum gradient that does not change the optimal value are analyzed.Finally,the direction of gradient descent with momentum is constructed to search for the deliverable aperture shape and the acceleration performance of the direction of gradient descent with momentum is studied.Research shows that the proposed method can use fewer apertures to obtain better treatment plan than the generic CG.(4)In addition,in order to overcome the disadvantage of slow convergence speed of the negative gradient descent direction of CG,a CG based on conjugate gradient descent direction is proposed.In the process of generating aperture shape,the proposed method uses the gradient in the aperture gradient map to construct the conjugate gradient descent direction to obtain deliverable aperture shape.The experimental results show that the proposed method can reduce the number of apertures required by the treatment plan,reduce the optimization time,and improve the dose distribution of each organ.Furthermore,in order to speed up the process of searching the aperture shape based on a single conjugate gradient descent direction,a CG based on the joint decision of two conjugate gradient descent directions is proposed.In the process of solving the pricing problem,two kinds of conjugate gradient descent directions are constructed according to the gradient in the aperture gradient map.Then,with the increase of the number of iterations,the weight of the two conjugate gradient descent directions in determining the search direction of the aperture shape is adjusted to obtain the deliverable aperture shape and add it to the treatment plan.Compared with the generic CG and the CG based on a single conjugate gradient descent direction,the objective function of the proposed method has the best descent speed,the algorithm speed is the fastest,and the quality of the treatment plan has been improved.