Noise-robust ML decomposition improves physical and biological dose accuracy in carbon-ion therapy planning DECT

A machine-learning method for dual-energy CT (DECT) elemental decomposition reduced water-equivalent range deviations by 0.4–1.4 mm and raised biological-dose gamma passing rates by up to 24.1% compared with parameterization approaches under clinically realistic noise.

• Four DECT elemental-decomposition methods were tested with 0%, 2%, and 5% image noise; the ML approach consistently yielded better accuracy for both physical and biological dose calculation.
• Biological dose modeling showed the Local Effect Model (LEM) had markedly superior noise robustness relative to the Linear Quadratic Model (LQM) and Microdosimetric Kinetic Model (MKM).
• This is a theoretical simulation study; clinical translation would require prospective validation with real patient DECT data.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.