Program Information
Computer-Aided Detection of Pathologically Enlarged Lymph Nodes On Non-Contrast CT in Cervical Cancer Patients for Low-Resource Settings
B Anderson*, C Cardenas , A Klopp , S Kry , J Johnson , J Ho , A Rao , J Yang , E Cressman , L Court , The University of Texas MD Anderson Cancer Center, Houston, TX
Presentations
SU-K-601-13 (Sunday, July 30, 2017) 4:00 PM - 6:00 PM Room: 601
Purpose: To create a program which automatically identifies pathologically enlarged lymph nodes on non-contrast simulation CT images obtained for radiation treatment planned for patients with cervical cancer. The program could be used in low-resource settings which commonly have CT simulation but not diagnostic imaging.
Methods: In a previous work, a region of interest that is likely to contain the nodal volumes relevant for cervical cancer was defined on a single patient CT. For our work, this region was deformed onto new patients using an in-house demons-based deformation software to define the search space. Hounsfield Unit thresholding was used within this masked region to generate a binary map of potential nodes. Edge detection and erosion filtering distinguished potential positive nodes from normal structures, such as the internal iliac arteries, descending aorta, and bladder. Regions on adjacent slices were then connected as a single potential node structure. Metrics were generated based on the shape and mean pixel values of the structures, and six different classification models were tested to differentiate the positive nodes from normal tissues. A cohort of 58 patients with physician-contoured external iliac and para-aortic nodes was used as a test-validation set. These patients contained 128 clinically significant nodes (greater than or equal to 1cm in axial extent).
Results: Classification model comparison led to the selection of the Adaboost ensemble model. Leave-one-out cross-validation of the 58 patients showed the ability to detect 75% of the clinically significant positive cervical cancer nodes with a false/true positive ratio of 8:1. The entire automated process takes ~20-minutes/patient.
Conclusion: Our computer-aided detection model can assist in the identification of metastatic nodal disease when contrast-enhanced CT and/or PET imaging is not readily available. By identifying these nodes, treatment fields can be modified to ensure that metastatic nodes are treated/boosted as required for curative radiation treatment.
Contact Email: