Lung cancer is associated with an overall 15% survival and is the most common cause of cancer death in the United States. Helical computer tomography (hCT) is very sensitive for detecting lung cancer by virtue of its cross-sectional perspective and ability to acquire volumetric data sets of the chest in single sequences of high resolution. Preliminary data show that lung cancers detected with hCT are frequently small, early stage lesions; however, indeterminate nodules are observed in 25% to 50% of screened individuals, the vast majority of which will be benign The exclusion of malignancy may require biopsy, surgery, or prolonged follow-up with CT. There is a compelling need to develop methods of accurate, non-invasive nodule characterization, possibly by follow-up with CT. There is a compelling need to develop methods of accurate, non-invasive nodule characterization, possibly by providing in vivo surrogates of the aberrant angiogenesis and altered cellular metabolism that are basic to neoplastic proliferation. Both contrast-enhanced CT and positron emission tomography (PET) are able to discriminate benign from malignant lung nodules. The former, based upon the principal that tumors enhance because of increased vascularity, has a high negative predictive value in lesions 7mm or greater, but is relatively non-specific, while PET using 2-[fluorine-18]fluoro-2-deoxy-D-glucose (FDG) is accurate, but reliable primarily for larger lesions of at least 15mm diameter. We propose to develop a combined CT/PET approach for nodule characterization, using image analysis with feature classification to investigate several potentially complementary imaging features to better discriminate benign and malignant lesions. From hCT, basic 2-D non- visual texture features; 3-D volume, shape, and surface features, shape, and surface features; and post-contrast enhancement patterns will be input to linear discriminant and non-linear neural net classifiers to determine the panel features that provides the greatest accuracy to determine the panel of features that provides the greatest accuracy. Semi-quantitative measures of metabolic activity in lung lesions will be calculated from FDG PET scans that have been optimally corrected for photon attenuation and partial volume effects by subject-specific anatomical templates created from hCT. The diagnostic performance of imaging features will be determined individually and in combination relative to diagnoses confirmed pathologically or by stable radiographic appearance over two or more years. In addition to predicting malignancy in vivo, the relationships between CT and PET features to morphometric and immunohistochemical preparations of lesions will be analyzed with a goal to identifying imaging patterns that are more predictive of tumor biology and clinical outcomes than is currently possible with classical descriptions of tumor size and stage. Our hypothesis is that the information derived from x-ray CT and FDG-PET can combine synergistically to improve the accuracy of nodule characterization for lesions both within and below the current threshold of accuracy of either modality.
Showing the most recent 10 out of 201 publications