MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

AutoMO: An Automated Multi-Objective Model for Reliably Predicting Lymph Node Metastasis in Head & Neck Cancer

Z Zhou*, M Dohopolski , L Chen , X Chen , S Jiang , D Sher , J Wang , UT Southwestern Medical Center, Dallas, TX

Presentations

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

Room: 225BCD

Purpose: Reliably classifying lymph node metastasis (LNM) plays an important role for diagnosis and treatment in head & neck (H&N) cancer. We aim to develop a novel PET/CT based automated multi-objective (AutoMO) model for LNM prediction.

Methods: The dataset includes 129 surgical H&N patients from 2009 to 2018 with total 543 lymph nodes (LNs). Benign and malignance status of each LN was determined from pathological reports. The training stage of AutoMO is to generate the Pareto-optimal model set. To avoid influence of negative features during the training stage, we perform feature selection and model training simultaneously. Then our iterative multi-objective immune algorithm is employed to maximize the sensitivity and specificity simultaneously and Pareto-optimal models are generated. The test stage consists of weight calculation and evidential reasoning (ER) based fusion. The weight is calculated based on the trade-off between sensitivity and specificity, as well as the area under the curve (AUC). The weights for the models whose sensitivity and specificity are extremely imbalanced are set as zero. Then the predicted probabilities from all the individual models are fused through the ER to obtain more reliable outcomes. Finally, the label with the maximal probability is considered as the final label.

Results: AutoMO was trained through 91 malignant nodes and 287 benign nodes, and tested on 39 malignant nodes and 126 benign nodes. AUC, accuracy, sensitivity, and specificity for AutoMO are 0.99, 0.93, 0.95, and 0.93, respectively. It outperforms traditional multi-objective model (MO) and conventional neural network (CNN) that achieve AUCs of 0.96, and 0.95, respectively.

Conclusion: We developed a novel AutoMO model for predicting LNM in H&N cancer. The results demonstrated that AutoMO outperformed MO and CNN.

Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by the US National Institutes of Health (R01 EB020366)

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email