A study on the radiomic correlation between CBCT and pCT scans based on modified 3D-RUnet image segmentation

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Locally advanced rectal cancer (LARC) staging relies on imaging.
• Cone-beam CT (CBCT) and positioning CT (pCT) are used for radiotherapy planning.
• Radiomics offers quantitative analysis of medical images.

Purpose of the Study:

• To investigate the correlation between CBCT and pCT radiomics features in LARC.
• To assess the potential of CBCT radiomics for tumor characterization and staging.
• To evaluate the substitutability of CBCT for pCT in radiomics analysis.

Main Methods:

• A deep learning model (CLA-UNet) was developed for tumor segmentation on CBCT and pCT.
• 103 LARC patients undergoing neoadjuvant chemoradiotherapy were included.
• Radiomics features were extracted and correlated between CBCT and pCT datasets.

Main Results:

• The CLA-UNet model achieved segmentation scores (DSC) of 0.792 for pCT and 0.672 for CBCT.
• 73 radiomics features showed high correlation (R > 0.9) between CBCT and pCT.
• Three of these features were linked to rectal cancer staging and prognosis.

Conclusions:

• An automated segmentation method for rectal cancer GTV was developed for both pCT and CBCT.
• CBCT-derived radiomics features hold significant value for research in rectal cancer.
• CBCT radiomics can potentially aid in staging and prognosis assessment for LARC.