Multivariate statistical process control for daily linear accelerator quality assurance: A framework for early detection and diagnosis.
Saved in:
| Title: | Multivariate statistical process control for daily linear accelerator quality assurance: A framework for early detection and diagnosis. |
|---|---|
| Authors: | Legrand, Christophe1,2 (AUTHOR) christophe.legrand@ico.unicancer.fr, Tiplica, Teodor2 (AUTHOR), Kobi, Abdessamad2 (AUTHOR) |
| Source: | Medical Physics. May2026, Vol. 53 Issue 5, p1-18. 18p. |
| Subjects: | Linear accelerators, Statistical process control, Radiotherapy, Fault diagnosis, Quality assurance, Radiation doses |
| Abstract: | Background: Statistical Process Control (SPC) has gained increasing interest in radiotherapy Quality Assurance (QA), particularly following recommendations from American Association of Physicists in Medicine Task Group 218. However, SPC is still predominantly applied via univariate charts such as Shewhart and Exponentially Weighted Moving Average (EWMA), even though many linac QA parameters (dose, symmetry, and flatness) are correlated and may drift together. Monitoring each parameter independently increases workload and overlooks the covariance structure, potentially reducing sensitivity to emerging faults and contributing to both false alarms and missed deviations. Multivariate Statistical Process Control (MSPC) techniques, such as Hotelling's T2 and Multivariate EWMA (MEWMA), address these limitations but remain underused in clinical practice. Purpose: This study aims to (1) provide a clear, practical framework for implementing T2 and MEWMA charts in radiotherapy machine QA, (2) compare their detection performance with standard univariate Shewhart and EWMA charts, and (3) demonstrate the diagnostic added value of variable‐level contribution analysis for identifying the root causes of out‐of‐control (OC) conditions. Methods: Daily QA measurements were collected over 168 days on a TrueBeam STx accelerator using a Daily QA3 device, yielding four dosimetric beam parameters: dose deviation, flatness, axial symmetry, and transverse symmetry. Univariate (I‐chart, EWMA) and multivariate (T2, MEWMA) charts were constructed following standard SPC methodology. Phase I data (90 observations) were validated for stationarity, independence, and normality using Kwiatkowski–Phillips–Schmidt–Shin, Augmented–Dickey–Fuller, Ljung–Box, Shapiro–Wilk, and several multivariate normality tests. The upper control limits for T2 and MEWMA were derived from the theoretical F distribution and χ2 approximation, respectively. Contribution analysis based on Cholesky decomposition was used to quantify variable‐level responsibility for each multivariate alarm. Detection performance was evaluated for three real clinical events: an abrupt dose anomaly, a progressive drift in symmetry/flatness, and a sudden monitor‐chamber failure. Results: Univariate charts detected the major deviations but required monitoring eight separate charts, increasing cognitive and operational burden. Both T2 and MEWMA charts detected all clinically relevant events identified by univariate charts, and often earlier. The MEWMA chart detected the onset of transverse symmetry degradation two measurements before the univariate EWMA and identified coordinated deviations across variables that were not yet individually out of control. T2 was particularly effective for abrupt shifts, mirroring univariate Shewhart performance but within a consolidated multivariate framework. Contribution analysis consistently identified the variable(s) driving each OC signal, providing clear diagnostic insight. Multivariate monitoring also revealed that the process never returned to an in‐control state between Events 2 and 3, a finding not apparent from the ±2% specification limit or univariate charts. This demonstrates the importance of accounting for covariance when interpreting QA data. Conclusions: Hotelling's T2 and MEWMA charts enhance early detection of deviations in linac QA by integrating the covariance structure among beam parameters and reducing the family‐wise false‐alarm rate inherent to multiple univariate charts. Their use streamlines workflow, improves diagnostic clarity through contribution analysis, and provides earlier warnings of progressive faults. MSPC represents a clinically valuable and operationally efficient extension to current radiotherapy QA practice, particularly as QA datasets become increasingly multidimensional. [ABSTRACT FROM AUTHOR] |
| Copyright of Medical Physics is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
| Abstract: | Background: Statistical Process Control (SPC) has gained increasing interest in radiotherapy Quality Assurance (QA), particularly following recommendations from American Association of Physicists in Medicine Task Group 218. However, SPC is still predominantly applied via univariate charts such as Shewhart and Exponentially Weighted Moving Average (EWMA), even though many linac QA parameters (dose, symmetry, and flatness) are correlated and may drift together. Monitoring each parameter independently increases workload and overlooks the covariance structure, potentially reducing sensitivity to emerging faults and contributing to both false alarms and missed deviations. Multivariate Statistical Process Control (MSPC) techniques, such as Hotelling's T2 and Multivariate EWMA (MEWMA), address these limitations but remain underused in clinical practice. Purpose: This study aims to (1) provide a clear, practical framework for implementing T2 and MEWMA charts in radiotherapy machine QA, (2) compare their detection performance with standard univariate Shewhart and EWMA charts, and (3) demonstrate the diagnostic added value of variable‐level contribution analysis for identifying the root causes of out‐of‐control (OC) conditions. Methods: Daily QA measurements were collected over 168 days on a TrueBeam STx accelerator using a Daily QA3 device, yielding four dosimetric beam parameters: dose deviation, flatness, axial symmetry, and transverse symmetry. Univariate (I‐chart, EWMA) and multivariate (T2, MEWMA) charts were constructed following standard SPC methodology. Phase I data (90 observations) were validated for stationarity, independence, and normality using Kwiatkowski–Phillips–Schmidt–Shin, Augmented–Dickey–Fuller, Ljung–Box, Shapiro–Wilk, and several multivariate normality tests. The upper control limits for T2 and MEWMA were derived from the theoretical F distribution and χ2 approximation, respectively. Contribution analysis based on Cholesky decomposition was used to quantify variable‐level responsibility for each multivariate alarm. Detection performance was evaluated for three real clinical events: an abrupt dose anomaly, a progressive drift in symmetry/flatness, and a sudden monitor‐chamber failure. Results: Univariate charts detected the major deviations but required monitoring eight separate charts, increasing cognitive and operational burden. Both T2 and MEWMA charts detected all clinically relevant events identified by univariate charts, and often earlier. The MEWMA chart detected the onset of transverse symmetry degradation two measurements before the univariate EWMA and identified coordinated deviations across variables that were not yet individually out of control. T2 was particularly effective for abrupt shifts, mirroring univariate Shewhart performance but within a consolidated multivariate framework. Contribution analysis consistently identified the variable(s) driving each OC signal, providing clear diagnostic insight. Multivariate monitoring also revealed that the process never returned to an in‐control state between Events 2 and 3, a finding not apparent from the ±2% specification limit or univariate charts. This demonstrates the importance of accounting for covariance when interpreting QA data. Conclusions: Hotelling's T2 and MEWMA charts enhance early detection of deviations in linac QA by integrating the covariance structure among beam parameters and reducing the family‐wise false‐alarm rate inherent to multiple univariate charts. Their use streamlines workflow, improves diagnostic clarity through contribution analysis, and provides earlier warnings of progressive faults. MSPC represents a clinically valuable and operationally efficient extension to current radiotherapy QA practice, particularly as QA datasets become increasingly multidimensional. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 00942405 |
| DOI: | 10.1002/mp.70511 |
