MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Development and Validation of a Machine Learning Predictive Model of IMRT Patient-Specific Quality Assurance Approval Using Gamma-Radiomics

C Yaly1, J Lizar2, P Santos3, A Colello Bruno1, G Viani1, J Pavoni1,2*, (1) Radiotherapy Department, Ribeirao Preto Medical School Hospital and Clinics, University of Sao Paulo, ,,BR, (2) Department of Physics, Faculty of Philosophy, Sciences and Letters at Ribeirao Preto - University of Sao Paulo, ,,BR, (3) Department of Psychology, Faculty of Philosophy, Sciences and Letters at Ribeirao Preto, University of Sao Paulo, ,,BR,

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: Gamma function is the standard method for comparing dose distribution, however, it suffered several criticisms. The use of radiomics was proposed as a very sensitivity methodology to evaluate gamma images calculated based on the field by field dose distribution measurements of patient-specific quality assurance (QA) of Intensity Modulated Radiation Therapy (IMRT) treatments. Thus, this study aims to apply the same methodology to evaluate radiomics of gamma distribution images based on composite dose distribution measurements and develop a patient-specific QA approval predictive model using machine learning (ML) algorithms.


Methods: radiomic features were extracted from 214 gamma distribution images obtained using a gamma criterion of 2%/2mm/15% threshold, and for the QA approval, 95% of the points should pass in the analyses. A data mining procedure was performed to reduce the number of features to be used in the model based on a t-test applied to evaluate if a significant difference could be found between the mean values of the radiomics features among the approved and reproved images. These features were used in a random forest ML model and its performance was evaluated and validated by the confusion matrices and ROC curves analyses.


Results: radiomics features showed statistical significance and were used in the ML model. The developed model presented a high accuracy and high ROC AUC for classification (0.85 and 0.93) and validation (0.81 and 0.84).


Conclusion: A high accurate and validated RF ML algorithms predictive model of the patient-specific QA test result based on the gamma-radiomics methodology of composite dose distribution measurements was developed.

Keywords

Quality Assurance, Intensity Modulation, Radiation Therapy

Taxonomy

TH- Radiation Dose Measurement Devices: Development (new technology and techniques)

Contact Email