MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Head Neck Cancer Locoregional Recurrence Prediction Using Delta-Radiomics Feature

K Wang1*, Z Zhou2, L Chen3, R Wang4, D Sher5, J Wang6, (1) UT Southwestern Medical Center, Dallas, TX, (2) University Of Central Missouri, Warrensburg, MO, (3) UT Southwestern Medical Center, Dallas, TX, (4) UT Southwestern Medical Center, Dallas, TX, (5) UT Southwestern Medical Center, Dallas, TX, (6) UT Southwestern Medical Center, Dallas, TX

Presentations

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

Room: AAPM ePoster Library

Purpose: Accurate identifying head and neck squamous cell cancer (HNSCC) patients at high risk of locoregional recurrence (LR) after radiotherapy is of great importance to guide physicians to develop personalized treatment strategies. This work aims to develop a multiple-classifier, multi-objective and multiple-modality (mCOM) model for HNSCC LR prediction, and to investigate whether the therapy induced changes characterized by radiomics features can help improve the accuracy of LR prediction.


Methods: The dataset includes 224 HNSCC patients received radiotherapy at our institute. Delta-radiomics features are extracted from PET/CT images acquired before and after radiation therapy, which are used in conjunction with radiomics features extracted from pre- and post-treatment images as model input. In mCOM, sensitivity and specificity are set as the optimization objectives; clinical parameter, CT and PET are considered as three modalities; and for each modality, three different classifiers, including support vector machine (SVM), logistic regression (LR), and discriminant analysis (DA), are utilized to obtain LR probability. An immune algorithm is employed to iteratively update feature selection vector, model parameters, weights of classifiers and pareto-optimal solution sets. After the training, for a testing sample, the output probabilities of solutions in the pareto-optimal solution set are fused using evidential reasoning (ER) method for each modality. The output probabilities of all the modalities are then fused as the final output probability by ER.


Results: Compared with other models constructed with single classifier, features from single modality, and pre-/post-treatment features only, the proposed mCOM model trained with delta-radiomics features achieved better AUC, accuracy, sensitivity, and specificity.


Conclusion: We extracted delta-radiomics features for predicting LR of HNSCC, and a mCOM model was proposed. Therapy induced change on radiomics features can help improve the LR prediction accuracy. In mCOM, the fusion of multiple classifiers and multiple modalities can improve the robustness of the predictive model.

Funding Support, Disclosures, and Conflict of Interest: This work is supported by Cancer Prevention and Research Institute of Texas (RP160661) and National Institutes of Health (R01 EB027898). The authors have no relevant conflict of interest to disclose.

Keywords

Image Analysis, Texture Analysis

Taxonomy

IM/TH- Image Analysis (Single Modality or Multi-Modality): Machine learning

Contact Email