Development of an intelligent linear regression model for dose estimation to patients during whole-body PET scan

Abstract

Positron emission tomography (PET) scans are vital in diagnosing cancer and neurological disorders but raise concerns due to exposure to ionising radiation. This research is focussed on the development of an intelligent regression model to investigate the effective radiation dose received by a patient during the whole-body PET scan. Our newly developed intelligent model refers to the application of artificial intelligence (AI) and machine learning (ML) techniques. Since underfitting and overfitting are basic issues of any ML model, data fitting methodology for developing intelligent regression is taken care of by implementing the least absolute shrinkage and selection operator (Lasso) and ridge regression. In order to have the comparative performance of our model, we have also applied support vector and decision tree-based ML techniques as regressors to predict radiation doses in whole-body PET scans, keeping patient safety in mind. By incorporating patient-specific data and imaging parameters, these models aim to accurately estimate radiation doses, thereby optimising imaging protocols and reducing unnecessary exposure risks. The study uses PET\({/}\)CT data from 2009 to 2012. The linearly-independent covariates applied in this model are age, weight, height, residence time and injected activity and the dependence variable is taken as the effective dose. Model performance is evaluated using root mean square error (RMSE). A systematic exploratory data analysis has been carried out to investigate data cleaning, missing information, scaling and normalisation. The top five organs such as the brain, stomach, kidney, adrenal and spleen are focussed to produce the traditional descriptive statistics of data summary. Least absolute shrinkage and selection operator (lasso) regression exhibit stable RMSE values for organ equivalent doses across genders, while substantial RMSE variations exist among different models and organs, suggesting sensitivity to specific organs and patient gender. Accurate dose estimation is pivotal for risk assessment and protocol optimisation. This study evidenced the need to improve radiation dosimetry for specific organs aiming at patient care and radiology practices by considering individualised factors in dose estimation methodologies to refine PET scan dose estimation methods.