Impact of Interfractional Error on Dosiomic Features

Area of Science:

• Medical Physics
• Radiation Oncology
• Radiomics

Background:

• Dose and image-based features (dosiomic features) are increasingly used in radiation oncology.
• Interfractional errors, or variations in patient positioning during treatment, can affect the accuracy and reliability of these features.
• Assessing the stability of dosiomic features under such errors is critical for their clinical application.

Purpose of the Study:

• To investigate the stability of dosiomic features under random interfractional error in lung cancer patients.
• To evaluate how the number of fractions influences the stability and values of dosiomic features.
• To identify dosiomic features that are sensitive to interfractional errors.

Main Methods:

• Dose distributions were simulated for 15 lung cancer patients with introduced interfractional errors (±3 mm).
• Dosiomic features were extracted from four regions of interest (GTV, PTV, heart, lungs) across different fraction number groups (2, 10, 20, 30).
• Stability was assessed using coefficient of variation (CV) and intraclass correlation coefficient (ICC); percent differences (PD) were calculated.

Main Results:

• Eleven of 93 dosiomic features showed high variability (CV ≥ 20%), with greatest variability in the Gross Tumor Volume (GTV).
• Feature stability decreased with fewer fractions; five features exhibited poor stability (ICC < 0.75).
• Mean dosiomic feature values remained consistent across fraction groups (ICC ≥ 0.9), though some showed >50% PD.

Conclusions:

• Certain dosiomic features demonstrate low stability under interfractional error, influenced by the total number of fractions.
• These findings highlight the need to consider interfractional error effects for reproducible radiomics in lung cancer.
• Further research should focus on robust dosiomic feature selection and error mitigation strategies.