Computerized detection of lung tumors in PET/CT images.

More and more hybrid PET/CT machines are being installed in medical centers across the country as combining computer tomography (CT) and positron emission tomography (PET) provides powerful and unique means in tumor diagnosis. Visual inspection of the images is a tedious and error-prone task and in many clinics the attenuation-uncorrected PET images are not examined by the physician, potentially missing an important source of information, especially for subtle tumors. We are developing a computer aided diagnosis software prototype that simultaneously processes the CT, attenuation-corrected PET, and attenuation-uncorrected PET volumes to detect tumors in the lungs. The system applies optimal thresholding and multiple gray-level thresholding with volume criterion to extract the lungs and to detect tumor candidates, respectively. A fuzzy logic based approach is used to reduce false-positive tumors. The remaining set of tumor candidates are ranked according to their likelihood of being actual tumors. We show the preliminary results of a retrospective evaluation of clinical PET/CT images.