Search

Purpose: While ART has been studied for years, the specific quantitative implementation details have not. In order for this new scheme of radiation therapy to reach its potential, an effective ART planning strategy capable of taking into account the dosedelivery history and patient's on‐treatment geometric model must be in place. This work performs a study of dynamic closed‐loop control algorithms for ART and demonstrates their utilities with data from phantom and clinical cases with on‐treatment cone‐beam CTimages.Method: In closed‐loop control, the controller is not run just once but repeatedly, each time receiving the current state of the system as its input. To meet the requirements of different clinical applications, two classes of algorithms are developed: those Adapting to Changing Geometry (ACG) and those Adapting to Geometry and DeliveredDose (AGDD). The former takes into account organ deformations found just before treatment. The latter optimizes the dose distribution accumulated over the entire course of treatment by adapting at each fraction not only to the anatomic information just before treatment but also to the dosedelivery history. The closed‐loop algorithms are showcased by phantom and clinical cases. Results: A comparison of the approaches with conventional open‐loop IMRT without adaptively incorporating feedback information indicates that closed‐loop ART may significantly improve the current practice. In both phantom and clinical studies, AGDD outperforms ACG algorithms in three aspects: target dose coverage, sensitive‐structure sparing, and steeper gradients around the tumor. Within the AGDD formalism, it is beneficial not to correct all the previous dosimetric errors at once right after the information is available but over a number of fractions until next set of feedback data is available. Conclusion:ART with closed‐loop dynamic algorithms substantially improve the dose distribution. In addition, the differing performance of the specific implementations shows that the algorithmic details matter.

Purpose: Current radiation treatment plans do not adapt to errors (and other deviations from the target dose distribution) in already delivered fractions. We assume that the dose actually delivered (including any errors) is known after its delivery. We devise a dynamic control algorithm compensating for these errors by adapting the plan for the current fraction. Method and Materials: We wrote MATLAB code that models treatment of a 2D phantom using the beamlet model. We used SNOPT (a commercial SQP optimization code) to select beamlet weights that minimize the weighted quadratic deviation from some desired dose. We also apply our algorithm to a head and neck case. Suppose the prescription dose is D* over N fractions; Di * is the dose planned for fraction i; and Di is the dose actually delivered in fraction i. Our baseline (algorithm) chooses for fraction i the feasible plan, Di *, that minimizes the weighted quadratic deviation from D*/N. Our adaptive algorithm selects for fraction k the feasible plan Dk * that minimizes the weighted quadratic deviation from D*(k/N)−D 1−…−Dk−1. Simulation determines the delivereddoseDi from the anticipated doseDi * by adding noise and incorporating setup error (translation and rotation of the patient) and tissue distortion caused by small organ motion. We compare the adaptive algorithm to the baseline and an algorithm with perfect foresight. For the algorithms, we compare the DVH of the cumulative doseD 1+…+DN and the margin needed to achieve a satisfactory cumulative delivereddose.Results: We achieved significant improvement in the objective function while delivering more dose to the tumor and less to the sensitive structures on our test case with setup errors of ±1cm and ±2°. Conclusion: This new paradigm of adaptive radiation therapy (ART) holds significant promise for us to improve the current radiation therapy.

Purpose: Current radiation treatment practice neither detects nor adapts to intrafraction organ motion beyond gating. We develop a simple optimization scheme for 4D IMRT which doesn't rely on gating and simulate its performance in the presence of the anticipated organ motion and unanticipated setup errors and tissue motion. Method and Materials: We wrote MATLAB code modeling treatment of a 2D phantom using the beamlet model. We also used geometry from a lung case. SNOPT (a commercial SQP optimization code) selects beamlet weights minimizing the weighted quadratic deviation from some desired dose. Suppose the beam‐on time is divided into N phases and the prescription dose is D*. For location r in phase i, let Di *(r) be the planned dose;Di (r) the actually delivered dose; and r=Ai (v) the anticipated location of voxel v. Our two baseline algorithms use static plans (Di *=Dj * for any phases i,j) and gating (Di *=0 for phases i≠1). In both cases we choose feasible D 1*+…+DN * minimizing the weighted quadratic deviation from D*. Our 4D algorithm selects (D 1 *,…,D N *)∈ arg min (D1∈ F,…, DN ∈ F) , Σ v α(v)(D1(A 1 (v))+…+DN(A N (v)) − D*(v)) 2 Simulation determines the delivered doseDi from the anticipated doseDi * by adding noise and incorporating setup error (translation and rotation of the patient) and tissue distortion caused by unanticipated small organ motion. For our algorithm and the baseline, we compare the DVH of the cumulative doseD 1+…+DN and the margin needed to achieve a satisfactory cumulative delivered dose.Results: We achieved significant improvement in the objective function (delivering more dose to the tumor and less to the organ) on our test case with 3cm motions. Conclusion: This new paradigm of 4D IMRT holds significant promise for improving the current radiation therapy.