As an advanced radiation therapy technique, intensity-modulated radiation therapy (IMRT) allows for a complex dose distribution, with sharp dose gradients often achieved at the edge of the planning target volume (PTV). As IMRT is able to create an intentionally inhomogeneous dose distribution, it has routinely been employed as a radiotherapy technique (1). Due to the complex features of the inverse planning and the complexity in the delivery process, safety and quality assurance (QA) are important (2). As an essential and important process, patient-specific QA analyzes and evaluates the deviation between the predicted dose and the delivered/measured dose. This process can be carried out by using film dosimetry, ionization chambers, electronic portal imaging device (EPID), two-dimensional (2D) array detector, three-dimensional (3D) dosimetric system, gel dosimetry, log file, etc. (3, 4). Introduced in a seminal work by Low et al. (5), the global gamma index method is the most widely used technique for comparison of dose distribution for patient-specific pretreatment QA in radiation therapy (5–7). The 2D or 3D gamma analysis could be chosen based on different equipment. As a 3D volume analysis, the 3D gamma analysis can be used by COMPASS, Delta⁴, ArcCheck, Dolphin, OCTAVIUS 4D, etc. As a plane analysis, the 2D gamma analysis is often used by array detector, just like EPID, log file, Mapcheck, etc. Compared with the 2D gamma analysis, the 3D gamma analysis is increasingly winning high favor among medical physicists because it can provide more comprehensive QA information. However, the 3D gamma analysis could make the Region of Interest (ROI) selection more complex.

In clinical practice, the results of the patient-specific QA should reveal the percentage dose errors (%DEs) of dose–volume histogram (DVH) metrics (i.e., if the results of the patient-specific QA was acceptable, the %DE should be ≤3%), which are the important parameters used by radiation oncologists (8). However, Nelms et al. (9), Stasi et al. (10), and Fredh et al. (11) have found that the conventional global gamma analysis has limited sensitivity to dose errors of DVH metrics. Recommended by the TG-218 (3), the conventional global gamma analysis for QA measurements and corresponding treatment plan was performed in absolute dose mode, adopting global normalization (often using the maximum planned dose point) and a dose threshold of 10%. In some studies, it has been argued recently that the percentage gamma passing rate (%GP) of individual structures is more sensitive to dose errors than that in the conventional gamma method (12, 13). The %GP of individual structures provides a greater insight into the dose delivered to target volume and organs at risk (OARs). Stasi et al. (10), Heilemann et al. (12), and Yi et al. (13) have adopted the %GP of individual structures instead of conventional global gamma. Because the %GP of individual structures can not only improve the weak correlation between the %GP and %DE but also provide spatial information of dose errors that are unavailable in conventional gamma analysis. However, this method is only suitable for plans for one body site with a simple anatomy just like the IMRT plan for cervical cancer. It is difficult to be promoted in head and neck plan verification due to multiple targets and OARs that cause inefficiencies and increased clinical workload due to an abundance of DVH parameters that must now be reviewed.

In some reports (14), the current IMRT QA practice has been fundamentally questioned due to the poor sensitivity and specificity of the global gamma index implementation. Under such circumstances, Stojadinovic et al. (15) and Yu et al. (6, 16) have proposed the divide and conquer (D&C) gamma method. In this method, the dose distribution is divided into distinct regions: i) a high-dose (HD) region within the 90% isodose, ii) a high-gradient (HG) region with doses ranging from 50% to 90% isodose, iii) a medium-dose (MD) region with doses ranging from 20% to 50% isodose, and iv) a low-dose (LD) region with doses ranging from 10% to 20% isodose. It has been indicated that the conventional gamma method includes almost all dose regions (≥10% isodose level). However, from a clinical perspective, not all dose regions are equally essential. By using the D&C gamma method, more information can be provided for validation without further increasing the time-intensive nature of IMRT QA. Hence, it has challenged the adequacy of the conventional approach to patient-specific QA analysis. However, these publications have neither illustrated the correlation between the QA results and clinical DVH metrics nor verified the sensitivity and specificity of the D&C gamma method.

In addition, the %GP has been explored in most studies, but the mean gamma index (μGI) has been accounted for in only a few publications. In fact, Stock et al. (17) have proposed the initial concept of achieving the μGI action limits (ALs) for the hot area (receiving 50% or more of the prescribed dose) as early as in 2005. Recently, Visser et al. (18) have completed the 3D dose verification by combining μGI evaluation with %DE. Few publications have explored the relationship between μGI and %DE or the impact of different evaluation regions on μGI.

In this study, a measurement-guided dose calculation system was used to divide the measured dose distribution into six isodose regions and one global clinically concerned region (PTV plus all OARs). By doing so, we aimed to investigate the optimal region where the correlation between %GP and %DE and between μGI and %DE can be improved. Besides, this study was also implemented to prove a new method that does not further increase the time-intensive nature of IMRT QA.

We selected 93 IMRT plans for nasopharyngeal cancer (NPC), including 61 original plans and 32 multileaf collimator (MLC) error-introduced test plans for retrospective analysis. These plans had been performed with measurement-based patient-specific QA. With ALs based on 3%/2 mm (global normalization) criteria recommended in American Association of Physicists in Medicine (AAPM) TG-218 (3), these IMRT QA plans were reanalyzed by using the different regions for evaluation.

In global gamma analysis, all points are normalized to one reference dose point where the maximum dose or isocenter dose is the most common choice (9). In general, errors are masked in the lower dose regions in the conventional global gamma because the low dose points are normalized to a high dose point (22). Furthermore, the TG-218 report has also implied that a low dose region (such as 10% of the prescription dose) can significantly affect the analysis (3). As a result, the conventional global gamma passing rate is often questioned regarding its limitations or clinical relevance (23, 24). Therefore, changing the evaluation region for the global gamma analysis (or changing the minimum threshold) is a preferred method because it can decrease the bias of the global gamma analysis.

In essence, the methodology proposed in our study computed the %GP and the μGI of reduced sample voxels of the dose distribution in different regions rather than that of all voxels. A conventional gamma analysis includes all voxels in the dose distribution (with dose values of 10% or greater). However, not all voxels have clinical significance in patient-specific QA. The targets and OARs are critical clinically relevant volume for making a final determination. The “whole” region for evaluation was explored with increasing dose level until it only contained the targets and OARs. The correlation between %GP, μGI, and %DE of various dosimetry metrics was used to assess the relevance of both %GP and μGI with clinically relevant observations.

From calculation equations 3–6, it is μGI rather than %GP that was able to more or less reflect the gamma value change of each evaluated point. The ROC analysis also demonstrated that the μGI in different regions was more sensitive than %GP. Therefore, it has implied that μGI reveals more clinically relevant problems compared with %GP.

In this study, both %GP and μGI results of seven regions for evaluation were found to have moderate correlation with the DVH metrics. This can be compared with the studies made by Zhen et al. (25) and Chan et al. (26). The μGI_whole was recommended to replace the conventional γ_10% and μGI_50% in the study. There are two reasons why the “whole” region is superior. One, from equations 3–6, both the spatial information and %DE were included in %GP and μGI. However, the dose difference information regarding specific structures was unavailable, which made %GP and μGI difficult to explore for assessing clinical implications (26). As the proposed “whole” region in our study only focused on the area of clinical concern, it can partially solve this problem. Moreover, Yu et al. (11) have demonstrated that different dose levels might include some other information such as the accuracy of TPS calculation in low-dose regions (10%–20% dose region), the accuracy of MLC modeling (50% dose region), and the expected information regarding the validity of the source model including beam energy, focal spot size, and collimator feature (high-dose regions). The “whole” region not only was the clinically concerned region but also provided the information in different dose levels. Although the sensitivities of μGI_60% and μGI_whole were similar, the “whole” region included more dose levels. Hence, it provided greater insight into the dose delivered to specific structures. According to the study by Cozzolino et al. (19) and Visser et al. (18), the DVH ALs should be set to 2%–5%. Therefore, we chose the DVH ALs of 3%. However, in areas with a sharp dose gradient and in metrics with small volumes, relatively high dose differences such as point dose (D₁) were observed. For results whose %DE exceeded DVH ALs, it was acceptable from the clinical point of view if the measured dose of the target volume was higher than the calculated dose or if the measured dose of OAR was lower than the calculated dose (27). The reason was that the results would not affect clinical efficiency. This view was proven by the %DE of the brain stem in Figure 5. The %DE of brain stem (D₁) of 32 cases was >3%, but most of the dose deviations were negative. Hence, this DVH ALs were too strict for OARs near the target volume such as the brainstem within or near the PTV in the NPC plan. The DVH ALs had an influence on defining the ALs of both %GP and μGI directly. So, if values determined for μGI_whole exceed the ALs, we need to check whether they have a clinical impact.

MLC errors mainly include random MLC positional errors and systematic positional errors. Random MLC positional errors have little dosimetric effect on IMRT plans (28). Steers and Fraass (23) concluded that the shifts in different directions may be an indication of different types of systematic errors arising from TPS model issues, machine issues, or other device issues. In addition, in our study, for the systematic MLC errors smaller than 0.8 mm, the positive rate of μGI_whole was higher compared with 0.75 mm systematic MLC errors in some publications (29), which were claimed to be undetected due to the poor resolution of array detectors. The author analyzed and concluded that the reason was that it was too strict to set the DVH ALs of 3% for the NPC plan. However, there is no consensus on the tolerance of the systematic MLC errors for patient QA. Numerous publications (30–32) have found that the systematic MLC errors up to 1 mm can produce a clinically relevant influence on the dose distribution. However, Rangel et al. (33) have suggested that systematic MLC errors need to be limited to 0.3 mm. Oliver et al. (34) have suggested that the systematic MLC errors should be within 0.63 mm to keep the PTV95 within 2%. In brief, μGI_whole can improve the detecting rate of small systematic MLC errors and the clinical effect of radiotherapy.

By gamma analysis, the verification results can be compressed into a single value (%GP or μGI). It can improve the efficiency of dose verification, which is usually important for clinics under a heavy workload. As proposed in this study, μGI_whole would not increase the clinical workload. Instead, it has a better correlation with DVH metrics than other commonly used evaluation metrics. Of course, these results also highlighted major limitations of our study. It was not a multicenter study and can potentially bias the analysis. In addition, these results may also have been affected by the specific choice of dosimeter, QA equipment, delivery system, and tumor sites (23). However, this study could be extended by stratifying the “whole” region by isodose level and anatomical site to provide more clinical information without increasing clinical work.

The empirical-type metrics, such as those in the patient-specific QA, may vary from institution to institution, since those metrics are based on local equipment, processes, and treatment types. Therefore, the conventional regions including “10%” and “50%” may not be suitable for all institutions. Different regions of the global gamma analysis have different diagnostic efficiencies. The results of μGI_whole can provide a new metric for the global gamma analysis.

The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding author.

This study was reviewed and approved by the Ethical and Scientific Committees of The First Affiliated Hospital of Chongqing Medical University (Chongqing, China). Written consent was obtained from the patient individual(s) for the publication of any potentially identifiable images or data included in this paper.

WL, YL, and XY conceived and designed the study and contributed to the analysis and interpretation of trials and literature and development and review of the article. HC and HZ collected the data, reviewed literature, and wrote the paper. WH prepared the figures and participated in the revision of literature. All authors contributed to the article and approved the submitted version.

This work was sponsored by Natural Science Foundation of Chongqing, China (cstc2021jcyj-msxmX0138).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.