MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Multiple Resolution Residual Network for Automatic Lung Tumor and Lymph Node Segmentation Using CT Images

H Um*, J Jiang , A Rimner , L Luo , J Deasy , M Thor , H Veeraraghavan , Memorial Sloan Kettering Cancer Center, New York, NY

Presentations

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

Room: 221CD

Purpose: To assess the performance of two deep network architectures in the segmentation of primary lung tumors and lymph nodes in patients with locally-advanced non-small cell lung cancer (LA-NSCLC).

Methods: The planning Computed Tomography (CT) scans of 241 LA-NSCLC patients treated at our institution were retrospectively analyzed for this study. The training and validation datasets consisted of 206 and 35 patients, respectively. All CTs were annotated with expert delineations of the primary tumors and lymph nodes as well as the heart and lungs. Two convolutional network architectures were evaluated: U-net and multiple resolution residual network (MRRN). The MRRN method simultaneously integrates features computed at multiple feature and resolution levels through a number of residually connected feature streams to compute the segmentation. Such a representation was used to increase the capacity of the network and to improve its ability to detect and segment small tumors and lymph nodes. We compared the MRRN method against the well-known U-net approach that uses a series of convolutions and pooling layers with skip connections to pass the information from the image at various resolutions for refining the segmentation. The networks were trained using 6758 2D images (size: 256x256, after cropping and resizing) and 3D segmentations were generated by combining these followed by morphological post-processing using hole filling, erosion, and dilation to obtain smooth contours. The segmentation accuracy compared to expert delineations was assessed using the Dice Similarity Coefficient (DSC).

Results: The detection rate of MRRN for the joint segmentation of primary tumors and lymph nodes was 60% and the median DSC was 0.70 (IQR: 0.59-0.72). The corresponding detection rate using U-net was 54% with a DSC of 0.58 (IQR: 0.46-0.64).

Conclusion: The results demonstrate we have successfully developed a deep learning approach for the automatic detection and segmentation of small-volume lung tumors and lymph nodes.

Funding Support, Disclosures, and Conflict of Interest: No related disclosures. Relevant financial activities outside the submitted work: Research grants from: Varian Medical Systems Boehringer Ingelheim Pfizer AstraZeneca Personal fees from: AstraZeneca Merck Research to Practice Cybrexa Non-Financial Support from: Philips/Elekta No stock, patent or other disclosures.

Keywords

CT, Segmentation, Lung

Taxonomy

IM/TH- image segmentation: CT

Contact Email