| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
|
|
|||||||||
|
|
| ||||||||||||||||||||||||||||||||||||||
Instrumentation & Data Analysis: Data Analysis & ManagementData Analysis & Management Posters |
1 Medical Physics, Sunnybrook, Toronto, Ontario, Canada; 2 Medical Biophysics; 3 Radiation Oncology, University of Toronto, Toronto, Ontario, Canada
1610
Objectives: To identify textural features and develop a method useful in automated segmentation of head and neck cancer (HNC) via quantitative characterization of Head and Neck tissues in co-registered FDG PET and CT images.
Methods: Abnormal and normal tissues of the head and neck were manually segmented from PET/CT images of 20 patients with HNC and 20 patients with lung cancer. Selected features from Spatial Gray-Level Dependence Matrices and Neighborhood Gray-Tone-Difference Matrices were used in characterization of these segmented ROIs. A decision tree (DT) based KNN classifier was developed to discriminate images of abnormal and normal tissues using multiple feature combinations. The area under the curve (AZ) of receiver operating characteristics (ROC) was used to evaluate the performance of features in comparison to an expert observer.
Results: The AZ of a DT based KNN classifier was 0.94. Sensitivity and specificity for normal and abnormal tissue classification were 88% and 99%, respectively.
Conclusions: Features such as PET Coarseness, PET Contrast and CT Busyness extracted from PET/CT images showed good discrimination performance. A DT-based KNN classifier demonstrated superior capability to classify tissues. The result can be used to develop an automated segmentation method which may lead to improvement in the accuracy of defining radiation targets.
Research Support: National Cancer Institute of Canada and the Ontario Cancer Research Network
| ||||||||||||||||||||||||||||||||||||||
