MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Prediction of Development of Rectovaginal Fistula Following Interstitial Brachytherapy for Advanced Cervix Cancer Using Machine Learning

Z Tian*, A Yen , Z Zhou , C Shen , K Albuquerque , B Hrycushko , UT Southwestern Medical Center, Dallas, TX

Presentations

(Monday, 7/30/2018) 9:30 AM - 10:00 AM

Room: Exhibit Hall | Forum 4

Purpose: Pelvic radiotherapy combined with interstitial brachytherapy is commonly used to treat patients with bulky, advanced cervical cancer. The proximity of invasive cancer to rectum and bladder and the high radiation dose needed to control the tumor can result in a rectovaginal fistula, which can be a devastating complication. We aim to identify patients at high risk for fistula formation and possibly prevent it from happening.

Methods: This IRB approved retrospective study included 32 cervical cancer patients treated at our institution, among which 6 have developed rectovaginal fistula after treatment. 20 clinical parameters were collected for each patient, including demographic parameters, tumor characteristics, health related parameters, dosimetric parameters and additional medical procedures. A support vector machine having a radial basis function as its kernel function was employed to build the prediction model. An iterative backward feature selection strategy was proposed to determine relevant parameters. To overcome data imbalance, the synthetic minority over-sampling technique (SMOTE) was used to generate artificial fistula cases for model training. K-fold cross-validation was employed to remove bias on feature selection and avoid overfitting in training.

Results: The feature selection process resulted in 9 relevant parameters: age, ethnicity, volume of HR_CTV, rectosigmoid involvement, rectosigmoid D2cc EQD2 dose, D1cc EQD2 dose, D0.1cc EQD2 dose, prior irradiation, and post-treatment biopsy. The AUC, accuracy, sensitivity, and specification achieved by our prediction model is 0.950, 87.7%, 93.1%, 86.4%, respectively, compared to 0.587, 84.0%, 31.3%, 96.2% without feature selection. Without SMOTE, we obtained 0.868 AUC, 76.2% accuracy, 21.0% sensitivity and 88.9% specificity when using the selected parameters, which demonstrated the effectiveness of SMOTE in overcoming data imbalance.

Conclusion: A prediction model of rectovaginal fistula development has been built for patients treated for bulky, advanced cervical cancer. This model may be clinically impactful pending refinement and validation in a larger series.

Keywords

HDR, Interstitial Brachytherapy, Modeling

Taxonomy

TH- Brachytherapy: GYN brachytherapy

Contact Email