PREDICTORY QUALITY CONTROL METHOD FOR DISTINGUISHING INSTALLATION UNCERTAINTY FROM MECHANICAL WEAR IN RADIOTHERAPY DEVICES

Abstract

This invention relates to a computer-based method for analyzing data obtained from external measurement equipment in the quality control processes of radiotherapy devices. The invention incorporates a filter mechanism that mathematically separates human-induced setup noise from the actual mechanical wear (active drift) of the device during testing, using time-weighted regression analysis (EWLS) and statistical significance parameters (p-value, R2). The method keeps the device's calibration memory updated with structural fracture tests, predicting when the machine will exceed its tolerance limits on a daily basis (Remaining Days - ETA), and preventing unnecessary false alarms, thus providing dynamic, proactive quality control management for clinics.

Description

1 TARIFF INSTALLATION UNCERTAINTY AND MECHANICAL ISSUES IN RADIOTHERAPY DEVICES. PREDICTIVE QUALITY CONTROL METHOD THAT DETECTS WEAR. Technical Area This invention is a medical linear accelerator 5 used in radiotherapy (radiation therapy) applications. It is concerned with managing the quality control (QA) processes of (LINAC) devices. More specifically... The invention identifies human-induced installation errors in quality control measurement data obtained from the hardware. (setup uncertainty), using time series and statistical algorithms to determine the device's true mechanical properties. mathematically separates wear from active drift and prevents the device from going outside its tolerance limits. 10 that generates dynamic maintenance alerts by predicting when it will be released (remaining days forecast). It involves a computer-based method. State of the Art Today, quality control (QA) processes of medical linear accelerator (LINAC) devices, data usually obtained from physical measuring instruments (phantoms, detector arrays, etc.) It is carried out through traditional quality control software that records data. The current technique is 15 In its known state, these software programs essentially operate on the logic of a "passive digital logbook". He is working. Examination of existing patent documents relating to the prior art reveals automated QA. These systems are seen. For example, patent number US11735309B2 belonging to Sun Nuclear. document 20 an automated quality assurance (QA) network and analysis framework for radiotherapy This has been explained. However, the system described in the patent is dependent on the device hardware and the manufacturer. It focuses on (hardware-coupled) data collection principles and external quality control. "Human-induced setup uncertainty" in the measurement data from the test, Statistical parameters of "mechanical wear and tear (active drift)" of the LINAC device. It does not offer a predictive filtering mechanism that differentiates. 25 2 Similarly, in Elekta's patent document numbered US10350438B2, during irradiation... Target tracking using quality indicators and automated machine data indicators. A system that provides tracking has been defined. This system monitors the dynamics during treatment. focusing on phantom setup during periodic device quality control (QA) tests. eliminating their variances with a mathematical decomposition filter (p-value, R-squared, etc.) or the device's 5 A decision that provides a predictive failure estimate (ETA) of when it will exceed tolerance limits. It does not suggest a specific structure. Current systems in the radiotherapy QA software market still perform data analysis in advance. a simple system based on defined fixed legal tolerance limits (e.g., ±2 mm geometric deviation) It operates on a "Pass / Fail" logic. This traditional approach has left 10 unresolved issues. It has the following structural problems: 1. Human-Induced Setup Error (Setup Noise) and Actual Machine Wear and Tear Inability to separate: Current software stores all data from the measuring device as purely absolute values. This is considered by the medical physicist as they place the measuring equipment on the treatment table. Misalignment errors of millimeters (installation uncertainty) can occur, resulting in a high risk of 15% deduction from the measurement results. This is reflected as variance (noise). Traditional systems attribute this sudden spike in value to an error by the physicist. whether it was caused by the installation or by a genuine malfunction of the LINAC hardware They cannot distinguish between them mathematically. This situation constantly causes "false alarms" in the clinic. and leads to unnecessary OEM callouts, causing financial burden on hospitals. It causes operational losses. 20 2. Reactive Detection and the Danger of "Blind Spots Between Tests": Current systems are purely predictive. It is not (predictive); it starts after a component physically exceeds its tolerance limit and the device triggers an alarm. They report the situation. Since statistical time series analysis is not applied, periodic tests are performed. Between these two points, it is not noticeable that the device is gradually drifting towards the tolerance limit. This The reactive structure puts patients at risk due to undetectable fatigue. 25 3. Calibration Amnesia (Forgetting Past Fatigue): Traditional QA systems, When a calibration is performed on the LINAC device, it resets the system's risk score to zero. This applies to mechanical components. 3 accelerated degradation rate between calibrations and structural degradation due to material fatigue. They cannot keep track of structural breaks by memorizing them. In conclusion, under the known state of the art, human setup noise in the device measurement data... Real machine wear can be estimated using exponential regression algorithms and statistical parameters. capable of differentiating, having calibration memory, and knowing when the device will go out of tolerance. a dynamic, closed-loop decision-making mechanism that can predict on a daily basis (Remaining Days - ETA) It is urgently needed. Purpose of the Invention The primary aim of this invention is to improve the existing radiotherapy quality control systems that exist in the known state of the art. (QA) software's passive, fixed-threshold, and purely reactive nature results in structural problems. By eliminating the disadvantages; hardware-independent, high-performance solutions for LINAC devices. The aim is to provide an accurate and closed-loop predictive quality control (QA) method. The underlying objectives of the invention and the main advantages it provides compared to the current state of the art are as follows:  Decoding and Identifying Human-Induced Setup Noise and Errors Preventing Alarms: The main objective of the invention is to prevent alarms from 15 external QA devices. daily random fluctuations in measurement data (for example, QA by medical physicists) setup errors such as alignment mistakes they make when placing the equipment on the treatment table The goal is to separate the uncertainties (of the system) from the actual mechanical wear (active drift) of the device. The developed decomposition filter is created using the standard deviation (σ) of the historical data. The installation noise reference profile is used. Regression applied to the time series 20 The statistical significance of the slope (p-value and R2) was determined by comparing the instantaneous measurement with this profile. Are the popping sounds just setup noise, or is it a hardware issue? Whether there is an active drift trend is precisely determined. Thus, in the clinic... Unnecessary false alarms and costly technical service calls are completely prevented.  Calculating the Real Machine Acceleration with Exponentially Weighted Least Squares (EWLS) Algorithm Capture: Unlike standard and unweighted regression models, the model applied in the invention With the EWLS method, the mathematical weight of the oldest measurements is exponentially reduced while the oldest measurements... More weight is given to current data. This allows us to avoid extreme setups from the past. 4 The errors do not distort the current calculation and the LINAC device's "current" actual drift. The trend is monitored with extreme sensitivity.  Generating Proactive Failure Prediction (Remaining Days - ETA): The invention relates to the use of equipment only. By solving the problem of alarms triggering in case of a malfunction or when a legal threshold is exceeded, we achieve "Zero Damage". It aims to operate on the (Zero-Harm) principle. The system calculates the drift slope as 5 (velocity) determines when the machine will exceed its tolerance limits on a daily basis. It predicts (ETA). This allows clinicians to anticipate when the device suddenly becomes unusable or proactive intervention before patients are exposed to suboptimal irradiation (e.g. (planned calibration) can be found.  Calibration Memory (10) with Structural Fracture Test Ensuring: Eliminating all past wear risks after calibration of existing systems. Solving the "calibration amnesia" problem that it had forgotten about is another innovative aspect of the invention. That is the purpose. Structural fracture testing (e.g., Chow Test) included in the system is applied to the device. after intervention, due to hardware material fatigue It detects whether there is an acceleration in wear rate or a fracture. A 15 If a break is detected, the old historical data queue is truncated, and only post-calibration data is available. The newly generated momentum is taken into account, thus providing continuous and reliable failure prediction. In conclusion, the invention involves everyday human setup rather than simply waiting for the threshold values ​​to be exceeded. a persistent machine that constantly filters out errors and only statistically verified ones. By generating ETA during validated active drift, it is used in radiotherapy quality control processes. It creates a completely new decision-making mechanism. Description of the Invention The invention utilizes data obtained from external measuring equipment in radiotherapy quality control processes. By analyzing, human-induced setup errors (setup noise) can be linked to mechanical wear and tear of the device. (Active drift) is a computer-based method for differentiating between different types of drift. This helps to better understand the invention. 25 For this purpose, the main steps and algorithmic flow that operate the system are detailed below: Data Acceptance and Activation Gateway (Rolling Window) Control The system includes radiotherapy device (LINAC) quality control equipment (e.g., daily QA). The process begins by accepting the incoming raw measurement data (phantoms or detector arrays). Before the predictive analytics engine is activated, the system undergoes an "Activation". The "gate" control is performed. In this step, a statistically significant noise profile is obtained. In order for it to be extracted, a predetermined minimum historical data pool is required (e.g., the last 90 days). It is checked whether at least 10 unique measurement days have occurred. If there is not enough data pool... The system only performs legal limit (absolute tolerance) checks; predictive analytics are performed if sufficient data is available. The motor is triggered. Derivation of Installation Noise Reference Profile (σ) After passing the activation gate, the system applies the relevant quality control parameter (e.g., Optical 10). Data from the last measurement period for Distance Indicator deviation (or Laser Alignment). It calculates the standard deviation (σ). This value is used daily by medical physicists with measuring instruments. It determines the "mathematical characteristic" of the random variance they generate in their setup. A high σ value indicates that the system is inherently chaotic in the nature of the test in question and that human intervention is required. It is interpreted and memorized as being susceptible to errors. 15 Time Series Weighting (EWLS Application) The system uses a standard to calculate the change in the device's daily performance (speed / velocity). Exponentially Weighted Least Squares (EWLS) instead of regression It uses the Squares algorithm. In this algorithmic step, the old QA measurements are analyzed. While the mathematical weight is reduced by an exponential decay factor, the most current 20 The highest statistical weight is given to the measurements. This allows us to analyze a large installation from months ago. The error is prevented from deviating from the current machine trend and the device's current drift speed is maintained. It is isolated with high precision. Drift Significance Test After a slope (trendline) is created by the EWLS algorithm, this trend is only 25 to prove whether it is random noise or permanent hardware wear Statistical significance tests are applied. In this step, the probability of the trend deviating from zero is shown. 6 The p-value is a measure of the fidelity of data points to a given trend line. The coefficient (R2) is calculated. Segregation Filter and Dynamic Decision Matrix The calculated statistical parameters are based on the pre-established setup noise reference profile (σ). The final classification is made by comparing them in a dynamic decision matrix: 5  High Setup Noise Scenario: The current measurement value suddenly changes. Even if it increases, if p > 0.05 and the R2 value is low, the algorithm will reject it. the deviation is not a trend of machine wear, but rather consistent with the historical noise profile. It mathematically proves that there was a human error in the installation. In this case, a fault alarm will be triggered. It is cancelled and the user is notified of an installation variance error, thus avoiding unnecessary technical service. 10 The call is blocked.  Actual Equipment Wear (Active Drift) Scenario: Daily deviations in measurements Even if the value is small, if p ≤ 0.05 and the R2 value is high, the algorithm will analyze the data; It confirms that it is moving in a consistent direction and that hardware fatigue is occurring. System In this scenario, the device will exceed its tolerance limit on day 15, calculated based on the historical acceleration. It generates a proactive Remaining Days (ETA) alert based on this. Structural Break and Fatigue Detection (Calibration Memory) In a preferred configuration, after a physical intervention or calibration of the device The system's slopes before and after calibration are measured using a structural fracture test (e.g., Chow Test). It analyzes. If, due to material fatigue in the LINAC equipment, 20 after calibration... If an acceleration (break) in the decay rate is detected, the system will revert to the pre-calibration data. By resetting its weighting, it only considers the current post-fracture slope. This step allows the device to... By preventing it from entering "calibration amnesia," it has become continuously adapted to machine fatigue. It provides a reliable ETA prediction.

Claims

7 REQUESTS 1. The invention relates to the mechanical properties of radiotherapy devices from quality control measurement data. a computer-based system that separates wear and tear from human-induced installation uncertainty The method and its characteristic is;  Acceptance of time series measurement data for at least one quality control parameter, 5  Based on the statistical distribution of accepted measurement data from the past period Creation of an installation noise reference profile (Sigma),  Time-weighted regression analysis is applied to the time series data in question to analyze the device's performance. Calculation of the working slope,  The statistical significance parameters of the calculated slope are based on the pre-established setup 10 Implementing a decomposition filter where the noise is compared with a reference profile,  The measurement, according to the output of the separation filter in question, indicates a location with high installation noise. classification as a condition or condition involving actual device wear and tear. Based on the results, it includes the steps to generate a dynamic device status report.

2. It is a method according to Claim 1, and its characteristic is that it uses time-weighted regression analysis, based on the most current 15 giving higher mathematical weight to the measurements and exponentially weighting the old measurements. with the Exponentially Weighted Least Squares (EWLS) algorithm which reduces it by It is the realization of.

3. A method according to claim 1 or 2, characterized by the application of a parsing filter. The statistical significance parameters used in this step, p-value (p-20 The values ​​are (value) and the coefficient of determination (R-squared).

4. It is a method according to claim 3, and its characteristic is that the p-value in the classification step is less than 0.

05. If the value is large and the R-squared value is low, the measurement is based on a high setup. It is classified as setup noise if the p-value is equal to or less than 0.

05. If this is the case and the R-squared value is high, then the actual device wear measured is 25. It is classified as (Active Drift).

5. A method that meets any of the above requirements, and whose characteristic is that the measurement is accurate. If classified as device wear, the device will exceed its tolerance limit. The process involves calculating the remaining time (ETA) and generating a proactive calibration notification. 8 6. A method according to any of the above requirements, and its characteristic is; the device Following calibration, a material fatigue issue occurred at the operating inclination of the device. Checking for structural fracture using a structural fracture test, fracture If detected, pre-calibration historical data is excluded from the calculation. Only the current acceleration after calibration is taken into account. 5 7. It is a method according to Claim 6, and its characteristic is that the structural fracture test mentioned is the Chow Test. It is the fact that.

8. It is a method according to claim 1, characterized by the acceptance of time series measurement data. Before that step, the data must be able to form a statistically meaningful noise profile. 10 whether it includes data points for a predetermined minimum number of days. It is the presence of a controlling activation gate step.