Abstract for presentation at 38th Annual Scientific Meeting of the Australian and New Zealand Society of Nuclear Medicine 2008

Automatic Lung Tumor Segmentation from PET Images Using K-Means Clustering

  • Mrs Deepti Rameswary, Centre for Medical Radiation Physics, University of Wollongong & Department of Nuclear Medicine, Wollongong Hospital, Australia
  • Dr Steven Harvey, Department of Nuclear Medicine, Wollongong Hospital, Australia
  • Prof Brian Hutton, Institute of Nuclear Medicine, University College London, United Kingdom
  • Prof Anatoly Rosenfeld, Centre for Medical Radiation Physics, University of Wollongong, Australia

    • A K-means based segmentation method for automatic lung tumor delineation from Positron Emission Tomography (PET) images is being presented in this work. The usefulness of PET in oncology depends greatly on the automatic detection and delineation of tumor regions from F-18 FDG PET images. Although a number of segmentation methods are available for medical images, PET image segmentation faces a number of challenges because of the poor spatial resolution of the emission images. This work presents a novel method of segmentation based on K-means clustering for the automatic detection and delineation of lung tumors from PET images. The algorithm performs segmentation of the lung images by using binary splitting along with K-means clustering. Among the segments identified, a lung tumor is found to be the segment that converges first, leaving the other segments classified as the background. This simple segmentation technique was found to extract tumors effectively from the lung PET images chosen for the experiments. Results obtained from the present analysis have also been compared with other standard segmentation methods. The proposed method will perform tumor delineation from PET images without the help of the counterpart CT images and without any prior knowledge of the tumor.

    Conference Organiser - ICMS Pty Ltd