MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Machine Learning Models in Predicting Pathological Complete Response After Neo-Adjuvant Chemoradiation for Locally Advanced Rectal Cancer

Y Zhang1,2*, M diMayorca1,3, L Shi4, X Sun4, S Jabbour1, Y Zhang1, N Yue1, K Nie1, (1) Department of Radiation Oncology, Rutgers Cancer Institute of New Jersey, New Brunswick, NJ (2) Department of Radiology, University of California,Irvine, CA, (3) Columbia University, New York (4) Department of Radiation Oncology, Sir Run Run Shaw Hospital, Zhejiang Univ., Hangzhou, ,CN,

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: To investigate the effectiveness using various machine learning models in predicting the neoadjuvant chemoradiation therapy (nCRT) response for patients with locally advanced rectal cancer (LARC).

Methods: 95 locally advanced rectal cancer (LARC) patients were included, all with pre-radiation (pre-RT) and mid-radiation (mid-RT) multi-parametric MRI sets as T1, T2, dynamic-contrast-enhanced (DCE) and diffusion weighted imaging (DWI) sequences. A total of 10940 radiomics features were extracted and collected. Sequential feature selection process was utilized to select the most robust predictors by repeating the process 1000 times. Finally, 9 classifiers were tested for predicting pathological complete response (pCR), which included: (1) conventional linear regression, (2) decision tree, (3) Gaussian naïve Bayes, (4) kernel naïve Bayes, (5) linear support vector machine (SVM), (6) quadratic SVM, (7) cubic SVM, (8) Gaussian SVM, and (9) k-nearest neighbor. The performance of the total 27 models was built, using pre-nCRT, mid-nCRT, and combined pre-nCRT + delta changes, and was evaluated using area under the receiver operating characteristics curves (AUC) with 4-fold validation.

Results: All machine learning based models out-performed conventional statistical method as linear regression, in predicting treatment outcome. The Gaussian SVM showed as the best model, with the highest AUC of 0.83 using combined time-points image information. By incorporating response pattern (as delta radiomics), it improved the prediction of pCR with higher AUC over pre-nCRT radiomics.

Conclusion: By combining pre-treatment and early response pattern, statistical models using machine learning methods may be rendered for final pathological response prediction.

Download ePoster [PDF]

Keywords

MRI, Image-guided Therapy

Taxonomy

IM/TH- Informatics: Informatics in Therapy (general)

Contact Email