2023 Fiscal Year Final Research Report
Development of unbiased evaluation method of predicted dose distribution by the event-mixing technique
| Project/Area Number |
21K07565
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 52040:Radiological sciences-related
|
|---|
| Research Institution | Teikyo University (2023)
Hiroshima University (2021-2022) |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2021-04-01 – 2024-03-31
|
| Keywords | イベントミキシング / 予測精度 / 品質管理 / 品質保証 / 放射線計測 / 医学物理 / 放射線治療 |
|---|
| Outline of Final Research Achievements |
This study developed an unbiased method to evaluate the predicted gamma passing rate (agreement of measured and simulated dose distributions) by introducing the best and worst limits of evaluation metrics. Specifically, the worst limit was estimated by applying the event-mixing technique that was originally developed in the particle and nuclear physics experiments to estimate the continuum background in invariant-mass spectroscopy to the pairs of measured and predicted gamma passing rates. Normalized achievement scores were defined for four different evaluation metrics (standard deviation, correlation coefficient, mean absolute error, and mean squared error) and confirmed their consistency. We also estimated the alert frequency for the cases requiring verification measurement, as a function of achieved precision. This alert frequency helps give a specific goal of predicting the performance to achieve.
|
| Free Research Field |
医学物理学
|
| Academic Significance and Societal Importance of the Research Achievements |
本研究では放射線治療の精度検証結果の予測精度を、素粒子原子核実験で開発された event-mixing 法を用いて評価する全く新しい手法を開発した。近年の人工知能関連技術の爆発的な普及により様々な物が生成され、その精度や完成度が急速に向上しているが、本研究は予測または生成された物の精度を評価する際に、予測対象の精度を差し引いて、予測モデルそのものの能力をバイアスをかけずに評価する方法を与えるものである。本研究で扱った予測対象は数値のみであるが、この考え方を一般化することにより数値以外の予測対象を扱う際の予測能力の定量的評価が可能になることが期待される。
|
