Click here to


Are you sure ?

Yes, do it No, cancel

A Reliable Multi-Classifier Multi-Objective Model for Predicting Gene Mutation in Clear Cell Renal Cell Carcinoma

X Chen1*, Z Zhou2 , R Hannan2 , K Thomas3 , P Kapur2 , J Brugarolas2 , I Pedrosa2 , J Wang2 , (1) Xi'an Jiaotong University, Xi'an, Shaanxi, China (2) The University of Texas Southwestern Medical Ctr, Dallas, TX (3) Weill Cornell Medicine, New York, NY


(Sunday, 7/14/2019) 5:00 PM - 6:00 PM

Room: 225BCD

Purpose: To predict gene mutations of mutations of BAP1, PBRM1, and VHL in clear cell renal cell carcinoma (ccRCC) using a reliable multi-classifier multi-objective (MCMO) model.

Methods: In this study, we used radiomics features from contrast enhanced CT images of 57 patients from our institution and the Cancer Imaging Archive (TCIA), where exome sequencing including information on VHL, PBRM1, and BAP1 genes was available. We proposed a MCMO radiogenomics model that predicts VHL, PBRM1, and BAP1 gene mutations in ccRCC using quantitative CT feature set. To obtain more reliable prediction results, similarity-based sensitivity and specificity were defined and considered as the two objective functions simultaneously during training. To take advantage of different classifiers (six classifiers including support vector machine, logistic regression, discriminant analysis, decision tree, K-nearest-neighbor, and naive Bayesian), the evidential reasoning (ER) approach was used for fusing the output of each classifier. Additionally, a new similarity-based multi-objective optimization algorithm (SMO) was developed to train the MCMO model.

Results: Using the proposed MCMO model, we achieved a predictive area under the receiver operating characteristic curve (AUC) over 0.85 for VHL, PBRM1 and BAP1 genes with balanced sensitivity and specificity. Furthermore, MCMO outperformed all the individual classifiers, and yielded more reliable results than other optimization algorithms and commonly used fusion strategies.

Conclusion: We developed a MCMO classifier for predicting gene mutations in ccRCC using quantitative CT feature set. Compared to single classifiers, multi-classifiers fused through ER can fully use information extracted by different classifiers. MCMO was trained by a developed SMO algorithm and can greatly improve prediction accuracy and reliability. In MCMO, the concept of reliable outcome prediction was proposed and applied to the optimization procedure, generating more reliable results.

Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by the National Natural Science Foundation of China 61401349 amd the American Cancer Society ACS-IRG-02-196.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email