Direct aperture optimization using an inverse form of back‐projection

Xiaofeng Zhu, Timothy J. Cullip, Gregg Tracton, Xiaoli Tang, Jun Lian, John Dooley, S ChangJournal of Applied Clinical Medical Physics 2014 · DOI

TL;DR

This paper introduces a novel Direct Aperture Optimization (DAO) method for Intensity-Modulated Radiation Therapy (IMRT) that leverages an "index" concept inspired by CT reconstruction. This approach effectively reduces treatment delivery time by optimizing multileaf collimator (MLC) segment shapes and weights without sacrificing dosimetric quality, showing significant time savings in clinical trials.

What Problem Does This Paper Address?

Intensity-Modulated Radiation Therapy (IMRT) is a powerful cancer treatment technique that aims to precisely deliver radiation to tumors while sparing surrounding healthy tissues. However, traditional IMRT planning involves a multi-step optimization process that can lead to complex treatment plans with a large number of segments. This complexity translates to longer treatment delivery times, which can introduce uncertainties due to patient motion and setup errors, and also impacts clinical workflow efficiency. While Direct Aperture Optimization (DAO) has emerged as a way to directly model MLC shapes and weights for faster delivery, existing DAO methods often rely on computationally expensive global objective functions (GFOs), especially when dealing with complex dose-volume histogram (DVH) constraints. This paper addresses the need for a faster, more efficient, and equally effective DAO method that avoids the computational burden of GFOs.

Methods

The core innovation of this paper is a new DAO approach that eschews a Global Objective Function (GFO). Instead, it employs an "index" concept, drawing an analogy to the multiplicative algebraic reconstruction technique (MART) used in Computed Tomography (CT) reconstruction. This index is defined as the ratio of the "optimal" (targeted) dose to the "current" computed dose for each voxel.

Here's a breakdown of the key methodological components:

  • Index Concept: For each voxel, an index is calculated based on its organ type (e.g., target vs. organ at risk) and the difference between the prescribed dose and the currently computed dose. For target volumes (PTV), the goal is to reach the prescribed dose, so the index drives the dose up if it's too low. For organs at risk (OAR), the goal is to keep the dose below a maximum allowed dose, so the index is formulated to encourage dose reduction.
  • Beamlet Index: These voxel indices are then aggregated to create a "beamlet index" for each individual beamlet. This beamlet index is used to adjust the beamlet's intensity, effectively guiding the optimization towards the desired dose distribution.
  • Segment Shape Optimization: The beamlet index is further utilized to optimize the shapes of MLC segments. If a beamlet passing near a leaf edge indicates an under-dose, the leaf opening is increased; if it indicates an over-dose, the opening is decreased. This is an iterative process, adjusting leaf positions by 1 mm increments.
  • Segment Weight Optimization: For fixed segment shapes, a "segment index" is calculated, representing the average beamlet index within that segment. This segment index is then used to update the segment's weight (beam-on time), aiming to deliver the required dose.
  • Adding New Segments: If further optimization is needed, the system allows for the addition of new segments. The cost of a new segment is calculated based on the potential improvement in the objective function, and a new segment shape is formed if it offers a significant benefit.
  • In-house Implementation: The entire system is implemented in C/C++ within the authors' in-house treatment planning system, PlanUNC.

Datasets

The paper describes a retrospective study involving 40 IMRT patients recruited between July and September 2011. These patients were treated with Siemens linacs. The study compared the new DAO approach against the conventional two-step IMRT optimization method. No new datasets were introduced; rather, existing patient data was used for evaluation.

Key Insights and Ideas

  • Inverse Form of Back-Projection: The core idea of using an "index" as an inverse form of back-projection from CT reconstruction is a clever adaptation. It allows for iterative refinement of the treatment plan by directly assessing the "error" (deviation from target dose) at each voxel and propagating that information to guide beamlet and segment adjustments.
  • Eliminating GFO Burden: The most significant insight is the successful implementation of DAO without a GFO. This bypasses the computational complexity and potential for local minima associated with GFOs, especially when dealing with non-convex DVH constraints, leading to faster planning.
  • Unified Index for Multiple Optimizations: The versatility of the index concept is remarkable. It's used not only for fluence map optimization but also for optimizing segment shapes, segment weights, and even for deciding when to add new segments. This unified approach simplifies the optimization framework.
  • Clinical Relevance of Speed: The paper strongly emphasizes that reducing treatment time is not just about workflow efficiency but also directly impacts treatment quality by minimizing uncertainties due to patient motion. This highlights a crucial practical consideration in radiotherapy.

Key Results

  • Reduced Segment Count: The DAO approach significantly reduced the average number of segments:
  • Prostate cases: from 90 to 40 segments (a 56% reduction).
  • Head and Neck cases: from 130 to 50 segments (a 62% reduction).
  • Reduced Treatment Time: This reduction in segments directly translated to faster treatment delivery:
  • Prostate cases: average time reduced from 13 minutes to 8 minutes (a 38% reduction).
  • Head and Neck cases: average time reduced from 15 minutes to 9 minutes (a 40% reduction).
  • Dosimetric Quality Maintained: Crucially, these improvements in efficiency were achieved without sacrificing dosimetric quality, as verified by isodose line distributions and DVH curves, and by meeting clinical customized goals.
  • Rotational IMRT Promise: The DAO approach also showed promise for optimizing rotational IMRT (rIMRT) with burst mode, with a head and neck case demonstrating potential for even faster delivery (estimated at around five minutes).

Limitations and Future Work

  • Retrospective Study: The study was retrospective, meaning it analyzed data from past treatments. Prospective studies would be needed to confirm these findings in real-time clinical practice.
  • Hardware Limitations: The authors mention that their current accelerators are not equipped with arc features, limiting their ability to deliver the optimized rotational IMRT plans. This suggests future work could involve testing these plans on machines with arc capabilities.
  • Development of Rotational IMRT: While promising, the rotational IMRT aspect is still under development. Further work is needed to refine the optimization points and ensure smooth delivery and hardware integration for burst mode rIMRT.
  • Source Code Availability: The authors offer their PlanUNC source code for research purposes, indicating a willingness to facilitate further development and validation by the community.

Summary updated

Abstract

Direct aperture optimization (DAO) has been used to produce high dosimetric quality intensity-modulated radiotherapy (IMRT) treatment plans with fast treatment delivery by directly modeling the multileaf collimator segment shapes and weights. To improve plan quality and reduce treatment time for our in-house treatment planning system, we implemented a new DAO approach without using a global objective function (GFO). An index concept is introduced as an inverse form of back-projection used in the CT multiplicative algebraic reconstruction technique (MART). The index, introduced for IMRT optimization in this work, is analogous to the multiplicand in MART. The index is defined as the ratio of the optima over the current. It is assigned to each voxel and beamlet to optimize the fluence map. The indices for beamlets and segments are used to optimize multileaf collimator (MLC) segment shapes and segment weights, respectively. Preliminary data show that without sacrificing dosimetric quality, the implementation of the DAO reduced average IMRT treatment time from 13 min to 8 min for the prostate, and from 15 min to 9 min for the head and neck using our in-house treatment planning system PlanUNC. The DAO approach has also shown promise in optimizing rotational IMRT with burst mode in a head and neck test case.
BibTeX
@article{zhu2014direct,
  author = {Xiaofeng Zhu and Timothy J. Cullip and Gregg Tracton and Xiaoli Tang and Jun Lian and John Dooley and S Chang},
  title = {{Direct aperture optimization using an inverse form of back‐projection}},
  journal = {Journal of Applied Clinical Medical Physics},
  year = {2014},
  doi = {10.1120/jacmp.v15i2.4545},
}
Figure