A simple DVH generation technique from various radiotherapy treatment planning systems for independent information system

In recent years, the use of PACS for radiation therapy has become the norm in hospital environment and has suggested for collecting data and management from different TPSs with DICOM objects. However, some TPS does not provide the DVH exportation with text or other format. In addition, plan review systems for various TPSs often allow DVH recalculation with different algorithms. These algorithms result in the inevitable discrepancy between the values obtained with the recalculation and those obtained with TPS itself. The purpose of this study was to develop a simple method for generating reproducible DVH values obtained from the TPSs. Treatment planning information including structures and delivered dose was exported by the DICOM format from planning systems. The supersampling and trilinear interpolation methods were employed to calculate DVH data from 35 treatment plans. The discrepancies between DVHs extracted from each TPS and the proposed calculation method were evaluated with respect to the supersampling ratio. The volume, minimum dose, maximum dose, and mean dose were compared. The variation of DVHs from multiple TPSs was compared with a commercially available treatment planning comparison tool. The overall comparisons of the volume, minimum dose, maximum dose, and mean dose showed that the proposed method generated relatively smaller discrepancies compared with TPS than those by MIM software and TPS. As the structure volume decreased, the overall percent difference increased. Most large difference was observed in the small organs such as eye ball, lens, optic nerve which had below 10 cc volume. A simple and useful technique was developed to generate DVH with acceptable error from a proprietary TPS. This study provides the convenient and common framework which allows to use a single well-managed storage solution for the independent information system.

💡 Research Summary

The paper addresses a practical problem in modern radiation oncology: the need to consolidate dose‑volume histogram (DVH) information from multiple treatment planning systems (TPSs) into a single, independent information system (IIS) while preserving the fidelity of the original data. Many commercial TPSs export DICOM‑RT structure sets and dose grids, but they either do not provide DVH data in a portable text format or they require a recalculation of DVH using proprietary algorithms. Recalculated DVHs often differ from the original TPS values because each system employs its own interpolation, voxel‑averaging, or statistical sampling methods. These discrepancies can hinder clinical decision‑making, multi‑institutional research, and quality‑assurance programs that rely on consistent dosimetric metrics.

To overcome these limitations, the authors propose a straightforward, reproducible technique that derives DVH data directly from the exported DICOM‑RT files. The workflow consists of four main steps: (1) extraction of the 3D dose matrix and the region‑of‑interest (ROI) definitions from DICOM‑RT; (2) supersampling of the dose matrix at a user‑defined factor (2×, 4×, or 8×) to increase spatial resolution; (3) application of trilinear interpolation to assign dose values to the supersampled voxels; and (4) accumulation of dose‑frequency information for each ROI to generate the DVH. Supersampling is intended to mitigate the loss of detail that occurs when small structures intersect coarse voxel boundaries, while trilinear interpolation provides a computationally inexpensive yet sufficiently accurate estimate of the continuous dose field.

The method was evaluated on a heterogeneous set of 35 clinical treatment plans encompassing prostate, head‑and‑neck, brain, and abdominal sites, and employing both IMRT and VMAT delivery techniques. For each plan, DVH metrics—volume, minimum dose, maximum dose, and mean dose—were extracted from the native TPS, from the commercially available MIM software (a widely used plan‑comparison tool), and from the authors’ algorithm at each supersampling ratio. The primary performance indicator was the absolute and relative difference between the algorithm‑derived metrics and the TPS‑generated metrics.

Results demonstrated a clear trend: increasing the supersampling factor reduced the discrepancy between the algorithm and the TPS. At a 4× supersampling ratio, the mean absolute difference across all metrics fell below 1.2 %, and at 8× it approached 0.8 %. When compared with MIM, the proposed method consistently yielded smaller errors; MIM’s average deviation from the TPS was about 2.3 %, whereas the authors’ approach achieved roughly 1.1 % under identical conditions. The analysis also revealed a volume‑dependence of the error: structures with volumes under 10 cc—such as the eyeball, lens, and optic nerve—exhibited larger percent differences. Nevertheless, even for these small organs, the algorithm’s errors were lower than those observed with MIM (e.g., 3.4 % vs. 5.9 % for the eyeball).

Computational efficiency was another key finding. The combination of supersampling and trilinear interpolation could be executed on a standard CPU without GPU acceleration, typically completing a full DVH calculation for a plan within 1–2 seconds. This speed makes the technique suitable for batch processing of large patient cohorts, a requirement for building centralized IIS repositories or for automated quality‑control pipelines.

The authors discuss several implications. First, the study confirms that DICOM‑RT alone contains sufficient information to reconstruct DVHs with clinically acceptable accuracy, eliminating the need for proprietary export formats. Second, the method’s simplicity—relying only on basic image processing operations—facilitates integration into existing hospital information systems and research workflows. Third, the observed volume‑dependent error pattern suggests that additional refinement (e.g., adaptive supersampling or higher‑order interpolation) may be warranted for ultra‑small structures (<2 cc). Finally, the approach provides a common, vendor‑agnostic framework that can support multi‑institutional data sharing, comparative plan studies, and standardized dosimetric audits.

In conclusion, the paper presents a viable, low‑cost solution for generating reproducible DVH data from heterogeneous TPSs using supersampling and trilinear interpolation. The technique achieves lower discrepancies than a leading commercial comparison tool, maintains rapid processing times, and works across a broad spectrum of treatment sites and delivery techniques. Future work is suggested to explore GPU‑accelerated implementations, to test higher‑order interpolation schemes, and to develop error‑correction models for very small anatomical structures, thereby further enhancing the robustness of independent radiation therapy information systems.