Abstract
2096
Objectives In Cancer Diagnosis from PET-CT images, the radiologists seek abnormal areas as isolated accumulations of FDG. However the color and contrast of regions sometimes depend on the color-mappings on the viewer set by an engineer. To avoid that, we have developed a method for detecting abnormal area and presenting them in a standardized color and density.
Methods The method consists of 2 phases. First the 3-dimetional outline of each organ is extracted on PET images automatically. Second, in each organ, “cliff” areas (CA) are detected. CA means an area which has a larger SUV than a organ-specific value and also shows a sudden rise of SUV. Then the CA is extended to neighboring regions which have larger SUV than the minimum SUV in the original CA. To verify the accuracy of algorithm, we compare the algorithm result in lung and mediastinum with the report written by 2 experienced radiologist (over 10 years). The study included 26 patients (F:M 4:22; age range 36-89, 72 ± 12 years) who have cancer(s) in lung or mediastinum. Radiologist pointed out 49 areas.
Results Algorithm can detect 47 areas as abnormal in 49 ‘true’ positives. But 2 area is detected as in another organ. False negative: 2 (all in mediastinum), False positive: 50 areas (18 in thoracic diaphragm, 14 in mediastinum, 9 in hilum). Sensitivity is 95.9%, PPV is 48.5%. Specificity of organs: 52.2%(left lung), 44.0%(right lung), 48.6%(mediastinum), so there is no difference between organs.
Conclusions The above algorithm could detect all abnormal areas involving comparatively-low SUV areas. More precise detection of outline of organs is expected
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.