Meeting Banner
Abstract #3038

Diameter-Independent Computer-Aided Classification of Mesorectal Lymph Nodes in Rectal Cancer: Preliminary Results

Wael Shabana1, Dilkash Kajal1, Rebecca E. Thornhill1

1Medical Imaging, The Ottawa Hospital, Ottawa, Ontario, Canada

The clinical MRI assessment of metastatic mesorectal lymph nodes plays an important role in predicting prognosis in rectal cancer. However, radiologic assessment offers limited sensitivity to predict the malignant potential of lymph nodes, particularly when conventional diameter criteria are used. We identified several diameter-independent shape and textural features from contrast enhanced T1-weighted MR images of mesorectal lymph nodes that were capable of delineating nodes with high malignant potential with 85% sensitivity, 78% specificity and 85% accuracy. With further optimization, these advanced shape and textural features promise to provide improved sensitivity compared to conventional radiologic methods.

Abstract PDF E-Poster Video

Keywords

according accuracy aided appeared applied arrive assessed assessment available axes beyond bias border borders branch breath brown cancer candidate care cause cavities classification classifier clinical combined computed computer contour contrast criterion cross death deliberately delineate delineated depend depicting descriptors designated determined diameter directly discriminative disease divided eccentricity electronics enabled enhanced equally excluded experienced extracted facts fast feature features finding fold gray greater health heterogeneity homogeneity hospital identification identified illustrate included independent index institute intensity intermediate interpolated irregularity leading length lymph males malignant manually masses materials medial median metastatic morphologic morphological network neural node nodes others patients patterns performance plot potential predictors preliminary procedure process prognosis programs promises quantitative radiologic radiologist radiology rectal related remaining repartitioning repeated repeating resolution retrospective risk scale scatter segmented sensitivity seven several shape slice smoothness society software specificity statistical studies subsequently sufficient system systematically technical textural third topology train transform trio utilized validation vibe volumetric wavelet years