Virtual Contrast-Enhanced CT Images for Cervical Cancer Radiotherapy

Figshare hosts a research document detailing a deep learning model for generating virtual contrast-enhanced CT images from noncontrast CT scans for 210 cervical cancer patients. The study evaluates the model's performance using metrics like MSE, PSNR, UQI, SSIM, and Dice similarity coefficient for target volume delineation in radiotherapy. The document reports results including a Dice coefficient of 0.95 and subjective image quality scores.

Use Cases

• Train an nnTransUNet model to synthesize virtual contrast-enhanced CT images from noncontrast CT scans for cervical cancer patients.
• Evaluate the feasibility of virtual contrast-enhanced CT for radiotherapy dose calculation by comparing CT value distributions in blood vessels and bone marrow.
• Assess target volume delineation accuracy by analyzing the Dice similarity coefficient and Hausdorff distance metrics for tumor contours.
• Compare image quality between virtual and real contrast-enhanced CT using subjective scores for artefacts, noise, and anatomical structure clarity.

Strengths

• Model performance is quantified with specific metrics: MSE of 0.958, PSNR of 48.332, UQI of 0.997, and SSIM of 0.976.
• Clinical validation involves data from 210 cervical cancer patients who underwent both noncontrast and contrast-enhanced CT scans.
• The study provides a direct comparison of dose distributions, reporting a relative dose difference in the target area of less than 2%.

Limitations

• The dataset is a 25.0 KB DOCX document containing study results, not the underlying image data or a structured tabular dataset.
• Clinical scope is limited to 210 patients with cervical cancer, which may limit generalizability to other cancer types or larger populations.
• Image quality assessment includes a subjective score for anatomical structure clarity that was slightly lower (3.7 points) than real contrast-enhanced CT (4 points).

Provenance

Source

Figshare, authored by Yukai Chen.

Collection Method

Research study using an nnTransUNet deep learning model on CT scans from 210 cervical cancer patients.

Freshness

Last updated on 2026-03-23.