MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All Sessions
  4. All Joint Imaging-Therapy Scientific Session
  5. Multi-Modality Convolutional Neural Network for Lymph Node Metastasis Prediction in Head and Neck Cancer

Multi-Modality Convolutional Neural Network for Lymph Node Metastasis Prediction in Head and Neck Cancer

L Chen1*, Z Zhou2 , D Sher3 , Q Zhang4 , J Shah5 , N Pham6 , S Jiang7 , J Wang8 , (1) UT Southwestern Medical Center, Dallas, Texas, (2) UT Southwestern Medical Center, Dallas, Texas, (3) UT Southwestern Medical Center, Dallas, Texas, (4) State Key Lab of Biotherapy and Cancer Center, West China Hospital, Sichuan, Chengdu, ,(5) UT Southwestern Medical Center, Dallas, Texas, (6) UT Southwestern Medical Center, Dallas, Texas, (7) UT Southwestern Medical Center, Dallas, TX, (8) UT Southwestern Medical Center, Dallas, TX

Presentations

(Thursday, 8/2/2018) 7:30 AM - 9:30 AM

Room: Davidson Ballroom B

Purpose: Accurate prediction of lymph node metastasis (LNM) status is critical in treatment planning and management for head and neck cancer patients. Cervical LNM status is routinely evaluated on PET and CT for target delineation during radiation therapy planning. The accuracy of LNM identification is strongly dependent on physicians’ experience. The purpose of this study is to develop an accurate and automatic deep learning based prediction model for LNM that integrates information from multiple modalities.

Methods: The study included 31 patients with head and neck cancer. Nodal status was reviewed by a radiation oncologist and a nuclear medicine radiologist for all patients. The lymph nodes for the first 21 patients were used for model training, including 53 involved nodes, 39 suspicious nodes, and 30 normal nodes. The trained model was tested on the remaining independent 10 patients with 13 involved nodes, 9 suspicious nodes, and 17 normal nodes. Since incorporating information from different modalities may enhance the predictive power, we constructed a convolutional neural network (CNN) model to predict LNM using both PET and CT images as inputs. Information from PET and CT is integrated by the first convolutional layer with different kernels. High-level features for LNM prediction are then extracted by alternatively arranged convolutional and max-pooling layers in the PET&CT-based CNN architecture. CNN models using PET or CT as input alone were also evaluated for the comparison.

Results: The accuracy and AUC values are 79% and 0.85 for PET-based and 77% and 0.87 for CT-based models, respectively. In contrast, the accuracy and AUC obtained by PET&CT-based model are 90% and 0.94.

Conclusion: The proposed PET&CT-based CNN algorithm achieves more accurate LNM prediction by integrating information from different imaging modalities.

Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by the American Cancer Society (ACS-IRG-02-196) and the US National Institutes of Health (R01 EB020366).

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email

Additional Links