Machine Learning Prediction of Excess Relative Risk for Radiation-Induced Solid Thyroid Cancer Among Nuclear Medicine...

Area of Science:

• Radiation Oncology
• Occupational Health
• Machine Learning

Background:

• Nuclear medicine professionals face chronic occupational exposure to iodine-131, increasing thyroid cancer risk.
• Existing risk prediction tools are not optimized for rapid, individualized assessment in occupational settings.
• This study explores machine learning as a surrogate for precise risk estimation.

Purpose of the Study:

• Develop and validate machine learning algorithms for predicting excess relative risk (ERR/Gy.RST) of radiation-induced thyroid cancer.
• Analyze the influence of dosimetric and demographic factors on predicted risk.
• Identify the optimal algorithm for occupational health surveillance.

Main Methods:

• A dataset of 4657 observations was adapted from Life Span Study data for occupational low-dose conditions.
• Five features (gender, age at exposure, current age, distance from source, cumulative dose) were used.
• Decision tree, random forest, and multilayer perceptron algorithms were trained.

Main Results:

• Cumulative absorbed dose showed a positive correlation with ERR/Gy.RST (r=0.63), while source distance had a strong inverse association (r=-0.38).
• The multilayer perceptron (MLP) algorithm achieved superior performance (R²=0.999, MSE=0.002, MAE=0.010).
• MLP significantly outperformed random forest and decision tree regressors.

Conclusions:

• The MLP algorithm serves as a high-fidelity surrogate for established risk projection tools in nuclear medicine.
• It enables computationally efficient, feature-integrated quantification of radiation-induced thyroid cancer risk.
• Machine learning offers a practical, scalable complement for individualized surveillance of radiation risk.