MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Automatic Liver and Tumor Segmentation Using Hierarchical Convolutional-Deconvolutional Neural Networks with Jaccard Distance

Y Yuan*, M Buckstein , Y Lo , The Mount Sinai Medical Center, New York, NY

Presentations

(Tuesday, 7/31/2018) 11:00 AM - 12:15 PM

Room: Davidson Ballroom B

Purpose: To investigate a fully automatic framework based on deep fully convolutional-deconvolutional neural networks (CDNN) for liver and liver tumor segmentation on abdominal CT images.

Methods: Automatic segmentation of liver and its tumors is an essential step for extracting quantitative imaging biomarkers for accurate tumor detection, diagnosis, prognosis and assessment of tumor response to treatment. MICCAI 2017 Liver Tumor Segmentation Challenge (LiTS) provides a common platform for comparing different automatic algorithms on contrast-enhanced abdominal CT images in 1) liver segmentation, 2) liver tumor segmentation, and 3) tumor burden estimation. We participated this challenge by developing a hierarchical CDNN framework. A simple CDNN model is firstly trained to provide a quick but coarse segmentation of the liver on the entire CT volume, then another CDNN is applied to the liver region for fine liver segmentation. At last, the segmented liver region, which is enhanced by histogram equalization, is employed as an additional input to the third CDNN for tumor segmentation. Jaccard distance is used as loss function when training CDNN models to eliminate the need of sample re-weighting.

Results: Our framework was trained using the 130 challenge training cases provided by LiTS. The evaluation on the 70 challenge testing cases resulted in a mean Dice-Similarity-Coefficient (DSC) of 0.963 (global 0.967) for liver segmentation, a mean DSC of 0.657 (global 0.82) for tumor segmentation, and a root-mean-square-error (RMSE) of 0.017 for tumor burden estimation, ranking our method in the first, fifth and third place among 350+ participants respectively (user: deepX). The entire segmentation took 33 seconds (average) per case.

Conclusion: Our method yielded top performance on a large-scale heterogeneous dataset. It can be used in various research and clinical tasks such as quantitative image analysis on liver cancer and automated liver contouring in radiotherapy planning, and can be generalized to other organs.

Funding Support, Disclosures, and Conflict of Interest: This work is supported by grant UL1TR001433 from the National Center for Advancing Translational Sciences (NCATS), NIH

Keywords

Not Applicable / None Entered.

Taxonomy

IM/TH- image segmentation: CT

Contact Email