Wencheng Shao , Liangyong Qu , Xin Lin , Weikang Yun , Ying Huang , Weihai Zhuo , Haikuan Liu
{"title":"Estimating patient-specific organ doses from head and abdominal CT scans via machine learning with optimized regulation strength and feature quantity","authors":"Wencheng Shao , Liangyong Qu , Xin Lin , Weikang Yun , Ying Huang , Weihai Zhuo , Haikuan Liu","doi":"10.1016/j.apradiso.2025.111788","DOIUrl":null,"url":null,"abstract":"<div><h3>Purpose</h3><div>This study aims to investigate estimation of patient-specific organ doses from CT scans via radiomics feature-based SVR models with training parameter optimization, and maximize SVR models’ predictive accuracy and robustness via fine-tuning regularization parameter and input feature quantities.</div></div><div><h3>Methods</h3><div>CT images from head and abdominal scans underwent processing using DeepViewer®, an auto-segmentation tool for defining regions of interest (ROIs) of their organs. Radiomics features were extracted from the CT data and ROIs. Benchmark organ doses were then calculated through Monte Carlo (MC) simulations. SVR models, utilizing these extracted radiomics features as inputs for model training, were employed to predict patient-specific organ doses from CT scans. The trained SVR models underwent optimization by adjusting parameters for the input radiomics feature quantity and regulation parameter, resulting in appropriate configurations for accurate patient-specific organ dose predictions.</div></div><div><h3>Results</h3><div>The C values of 5 and 10 have made the SVR models arrive at a saturation state for the head and abdominal organs. The SVR models’ MAPE and R<sup>2</sup> strongly depend on organ types. The appropriate parameters respectively are C = 5 or 10 coupled with input feature quantities of 50 for the brain and 200 for the left eye, right eye, left lens, and right lens. the appropriate parameters would be C = 5 or 10 accompanying input feature quantities of 80 for the bowel, 50 for the left kidney, right kidney, and 100 for the liver.</div></div><div><h3>Conclusions</h3><div>Performance optimization of selecting appropriate combinations of input feature quantity and regulation parameters can maximize the predictive accuracy and robustness of radiomics feature-based SVR models in the realm of patient-specific organ dose predictions from CT scans.</div></div>","PeriodicalId":8096,"journal":{"name":"Applied Radiation and Isotopes","volume":"220 ","pages":"Article 111788"},"PeriodicalIF":1.6000,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Applied Radiation and Isotopes","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0969804325001332","RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"CHEMISTRY, INORGANIC & NUCLEAR","Score":null,"Total":0}
引用次数: 0
Abstract
Purpose
This study aims to investigate estimation of patient-specific organ doses from CT scans via radiomics feature-based SVR models with training parameter optimization, and maximize SVR models’ predictive accuracy and robustness via fine-tuning regularization parameter and input feature quantities.
Methods
CT images from head and abdominal scans underwent processing using DeepViewer®, an auto-segmentation tool for defining regions of interest (ROIs) of their organs. Radiomics features were extracted from the CT data and ROIs. Benchmark organ doses were then calculated through Monte Carlo (MC) simulations. SVR models, utilizing these extracted radiomics features as inputs for model training, were employed to predict patient-specific organ doses from CT scans. The trained SVR models underwent optimization by adjusting parameters for the input radiomics feature quantity and regulation parameter, resulting in appropriate configurations for accurate patient-specific organ dose predictions.
Results
The C values of 5 and 10 have made the SVR models arrive at a saturation state for the head and abdominal organs. The SVR models’ MAPE and R2 strongly depend on organ types. The appropriate parameters respectively are C = 5 or 10 coupled with input feature quantities of 50 for the brain and 200 for the left eye, right eye, left lens, and right lens. the appropriate parameters would be C = 5 or 10 accompanying input feature quantities of 80 for the bowel, 50 for the left kidney, right kidney, and 100 for the liver.
Conclusions
Performance optimization of selecting appropriate combinations of input feature quantity and regulation parameters can maximize the predictive accuracy and robustness of radiomics feature-based SVR models in the realm of patient-specific organ dose predictions from CT scans.
期刊介绍:
Applied Radiation and Isotopes provides a high quality medium for the publication of substantial, original and scientific and technological papers on the development and peaceful application of nuclear, radiation and radionuclide techniques in chemistry, physics, biochemistry, biology, medicine, security, engineering and in the earth, planetary and environmental sciences, all including dosimetry. Nuclear techniques are defined in the broadest sense and both experimental and theoretical papers are welcome. They include the development and use of α- and β-particles, X-rays and γ-rays, neutrons and other nuclear particles and radiations from all sources, including radionuclides, synchrotron sources, cyclotrons and reactors and from the natural environment.
The journal aims to publish papers with significance to an international audience, containing substantial novelty and scientific impact. The Editors reserve the rights to reject, with or without external review, papers that do not meet these criteria.
Papers dealing with radiation processing, i.e., where radiation is used to bring about a biological, chemical or physical change in a material, should be directed to our sister journal Radiation Physics and Chemistry.