J Wu¹, X Xing², Q Wang³, T Wu⁴, Z Zhou⁵*, (1) Shaanxi Normal University, Xi'an, China (2) Shaanxi Normal University, Xi'an, China (3) Shaanxi Normal University, Xi'an, China (4) Hospital Of Stomatology, Xi'an Jiaotong University, Xi'an, China (5) University of Kansas Medical Center, Kansas City, KS

• 
  Z Zhou

PO-GePV-M-307 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: Distant Metastasis (DM) has become one of the most important prognostic factors regarding long-term survival in oropharyngeal cancers (OPCs). It is important to predict whether the cancer cells of patients have been metastasized, and the survival rate can be improved by making personalized treatment. We developed a new principal component analysis (PCA) based BRB (PCA-BRB) model to predict DM in OPCs.

Methods: In this study, there are 232 real clinical data (Princess Margaret Cancer Center) totally used. Since belief rule base (BRB) can integrate the clinical knowledge into the modelling in an interpretable way, it is used. However, when more attributes are used to build BRB based model, too many rules are generated, leading to higher model complexity. To overcome this issue, a new principal component analysis (PCA) method based BRB (PCA-BRB) is developed in this study, where PCA is used to reduce attribute dimension. To obtain more accurate results, differential evolutionary (DE) is used for model optimization. The final output is inferred by an evidential reasoning (ER) approach. In the experiments, we use accuracy, sensitivity, specificity, and area under the curve (AUC) as the evaluation index.

Results: The mean and standard deviation value of PCA-BRB on the accuracy, sensitivity, specificity, and AUC are 0.8258±0.0455, 0.8106±0.1136, 0.8182±0.0909, and 0.8951±0.0347, respectively, while BRB is 0.7681±0.0753, 0.6808±0.1868, 0.8073±0.1085, 0.8106±0.0492.

Conclusion: A new PCA-BRB model for predicting DM in oropharyngeal cancer was developed in this study. The experimental results on real patient dataset demonstrated that PCA-BRB outperformed BRB and other models as well.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email