Gender Medicine in Computed Tomography Radiomics Analysis to Predict Disease Progression in Liver Respectable Colorectal Cancer Patients

Abstract

Background

Gender medicine is an evolving discipline that examines how diseases manifest and progress differently in men and women. Tailoring medical therapies and diagnostic approaches can enhance patient outcomes. While radiomics is emerging as a promising tool in personalized medicine, few studies evaluate its role in gender medicine within radiology. In this context, our preliminary objective was to determine whether radiomic features could predict disease-free survival within 3 years after the last follow-up in patients with colorectal liver metastases, with an emphasis on gender differences.

Methods

The study analyzed preoperative CT scans of 196 patients from The Cancer Imaging Archive who underwent resection of colorectal cancer liver metastasis. Using the Pyradiomics library, we extracted 1316 features for each patient. We developed an analysis framework applied initially to the entire patient sample, then separately to male and female subsamples. This framework included: Volume of Interest (VOI) segmentation, handcrafted feature extraction and selection, detection of confounding patients, and training of ensemble classification models comprising five classifiers. Performance was assessed through 100 rounds of 10-fold cross-validation.

Results

The selected feature subsets for male and female subsamples showed no overlap. The ensemble model demonstrated a notable improvement in performance when trained on the female subsample (mean AUC of 80.5%) compared to the model trained on the entire dataset (mean AUC of 64.8%), while performance for the male subsample remained nearly unchanged.

Conclusion

Although further validation with a larger dataset and external confirmation is needed, these preliminary results suggest a meaningful impact of gender medicine in radiology.