MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Tumor Co-Segmentation in PET/CT Using Cross-DenseNet for Lung Cancer

X Zhao1 , X Tan2 , L Li1 , W Lu3 , S Tan1*, (1) Huazhong University of Science and Technology, Wuhan, Hubei, (2) Hunan University of Technology, Zhuzhou, Hunan, (3) Memorial Sloan Kettering Cancer Center, New York, NY

Presentations

(Tuesday, 7/31/2018) 10:00 AM - 10:30 AM

Room: Exhibit Hall | Forum 9

Purpose: CT imaging has a high spatial resolution but often fails at providing a clear boundary between a tumor and its surrounding normal soft tissues in non-small cell lung cancer (NSCLC). PET imaging is able to provide a high contrast of standardized uptake values (SUVs) in NSCLC but at a low spatial resolution. Automatic segmentation of a NSCLC tumor using a single image modality (PET or CT) is usually challenging. We proposed a novel bi-modality 3D deep-learning tumor co-segmentation method for NSCLC, to utilize both PET and CT information simultaneously.

Methods: The proposed model was constructed based on the 3D fully convolutional neural network (CNN) with dense connection (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Two parallel DenseNet branches was used to extracted features from PET and CT, respectively. To fuse features extracted from two different image modalities, a cross concatenation was designed at the end of each dense block between the two parallel branches. The prediction of the tumor was the output of the whole network through a softmax function. The effectiveness of the proposed method was validated on a dataset of 74 NSCLC patients (48 for training and 26 for testing). The ground truth was manually delineated by an experienced radiation oncologist using the complementary visual features of PET and CT. The segmentation accuracy was evaluated by Dice similarity index (DSI).

Results: The proposed method achieved an average DSI of 0.85, showing a significant improvement over methods solely using PET and CT. In addition, the training time of the proposed network was reduced by nearly three hours over those using a standard CNN method.

Conclusion: The results demonstrated that the designed network was effective and fast, and was capable of taking full advantages of both PET imaging and CT imaging simultaneously.

Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by National Natural Science Foundation of China (NNSFC), under Grant Nos. 61375018 and 61672253.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email