2020/05/01 by Haydee M. Linares Rosales, Gabriel Couture, Rosales, Haydee M. Linares +9
Agricultural and Biological Sciences · Medicine · Physics and Astronomy · #Advanced Radiotherapy Techniques #FOS: Physical sciences #Medical Physics (physics.med-ph) #Radiation Dose and Imaging #Radiation Effects and Dosimetry
paper · pdf · doi:10.48550/arxiv.2005.00398
openalex publication_date 2020/05/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Purpose: We sought to evaluate the feasibility of using machine learning\nalgorithms for multipoint plastic scintillator detector calibration in\nhigh-dose-rate brachytherapy. Methods: The dosimetry system consisted of an\noptimized 1-mm-core mPSD and a compact assembly of photomultiplier tubes\ncoupled with dichroic mirrors and filters. An 192Ir source was remotely\ncontrolled and sent to various positions in a homemade PMMA holder. Dose\nmeasurements covering a range of 0.5 to 12 cm of source displacement were\ncarried out according to TG-43 recommendations. Individual scintillator doses\nwere decoupled using a linear regression model, a random forest estimator, and\nartificial neural network algorithms. The performance of the different\nalgorithms was evaluated using different sample sizes and distances to the\nsource for the mPSD system calibration. Results: The decoupling methods'\ndeviations from the expected TG-43 dose generally remained below 20%. However,\nthe dose prediction with the three algorithms was accurate to within 7%\nrelative to the dose predicted by the TG-43 formalism for measurements\nperformed in the same range of distances used for calibration. The performance\nrandom forest was compromised when the predictions were done beyond the range\nof distances used for calibration. The dose prediction by the linear regression\nwas less influenced by the calibration conditions than random forest, but with\nmore significant deviations. The number of available measurements for training\npurposes influenced the random forest and neural network models the most. Their\naccuracy tended to converge toward deviation values close to 1% from a number\nof dwell positions greater than 100. Conclusions: In performing HDR\nbrachytherapy dose measurements with an optimized mPSD system, ML algorithms\nare good alternatives for precise dose reporting and treatment assessment.\n