Click here to

Session: Multi-Disciplinary: Image-Guided Treatment Response Assessment [Return to Session]

Predicting Pathological Complete Response Through Automatic Belief Rule Base in Gastric Cancer

Q Wang1*, J Wu2, Z Wang3, M Liu4, M Ma5, Z Zhou6, (1) Shaanxi Normal University, ,,(2) Shaanxi Normal University, ,,(3) Peking University Cancer Hospital & Institute, ,,CN, (4) Shaanxi Normal University, ,,(5) Shaanxi Normal University, ,,(6) University of Central Missouri, Warrensburg, MISSOURI


WE-IePD-TRACK 4-4 (Wednesday, 7/28/2021) 3:00 PM - 3:30 PM [Eastern Time (GMT-4)]

Purpose: Neoadjuvant chemotherapy (NAC) has been proved as an effective way to improve survival time in gastric cancer treatment. Accurate predicting pathological complete response (pCR) status after NAC plays an important role in making the personalized treatment plan. We aim to develop a new automated belief rule base (AutoBRB) model to predict pCR after NAC in gastric cancer.

Methods: Totally 55 patients’ clinical data (Peking Cancer Hospital, Beijing, China) are used in this study. Since belief rule base (BRB) can integrate clinical domain knowledge and be interpretable, it is used in this study. However, it is always hard to initialize the model parameters. As such, a new automatic BRB (AutoBRB) is developed in this study. In AutoBRB, the referential values, adopted to construct the rule base, are initialized through the information gain ratio, and the corresponding belief degrees are initialized through a new grid strategy. To obtain optimized model, a differential evolutionary algorithm is employed to train the model parameters, and an adaptive searching strategy is adopted to set the confidence threshold. Furthermore, the final output is inferred through an evidential reasoning approach where the belief degrees are obtained and used for making the decision. In the experiments, accuracy, sensitivity, specificity, and area under the curve (AUC) are used as the evaluation measures.

Results: The mean and standard deviation value of AutoBRB on the accuracy, sensitivity, specificity, and AUC are 0.8962±0.0562, 0.8530±0.1161, 0.9394±0.0683, and 0.9488±0.0345, respectively, while BRB is 0.7659±0.0560, 0.9046±0.0756, 0.6273±0.1081, 0.9179±0.0450.

Conclusion: A new AutoBRB model for predicting pCR after NAC in gastric cancer was developed in this study. The experimental results demonstrated that AutoBRB is superior to BRB and other available methods.



    Not Applicable / None Entered.


    Not Applicable / None Entered.

    Contact Email