# Ultrasonic radiomics-based nomogram for preoperative prediction of residual tumor in advanced epithelial ovarian cancer: a multicenter retrospective study

## TL;DR

This study develops a model combining ultrasound radiomic features and clinical data to predict residual tumor status in advanced ovarian cancer patients before surgery.

## Contribution

A novel nomogram integrating ultrasound radiomics and clinical variables for predicting residual tumor status in ovarian cancer is developed and validated.

## Key findings

- The model achieved an AUC of 0.82 and accuracy of 0.84 in external validation.
- Combining radiomic and clinical features outperformed models using either alone.
- The model showed strong generalizability for preoperative prediction of complete tumor resection.

## Abstract

To identify radiomic features extracted from ultrasound images and to develop and externally validate a comprehensive model that combines clinical data with ultrasound radiomics features to predict the residual tumor status in patients with advanced epithelial ovarian cancer (OC).

The study included 112 patients with advanced epithelial OC who underwent preoperative transvaginal ultrasound. Of these, 78 patients were assigned to the development cohort and 34 to the external validation cohort. Tumor contours were manually delineated as regions of interest (ROI) on the ultrasound images, and radiomic features were extracted. The final set of variables was identified using LASSO (least absolute shrinkage and selection operator) regression. Clinical features were also collected and incorporated into the model. A combination model integrating ultrasound radiomics and clinical variables was developed and externally validated. The performance of the predictive models was assessed.

A total of 1,561 radiomic features and 18 clinical features were extracted. The final model included 10 significant ultrasound radiomic variables and 4 clinical features. The comprehensive model outperformed models based on either clinical or radiomic features alone, achieving an accuracy of 0.84, a sensitivity of 0.80, a specificity of 0.75, a precision of 0.88, a positive predictive value of 0.81, a negative predictive value of 0.86, an F1-score of 0.78, and an AUC of 0.82 in the external validation set.

The comprehensive model, which integrated clinical and ultrasound radiomic features, exhibited strong performance and generalizability in preoperatively identifying patients likely to achieve complete resection of all visible disease.