Computed diagnosis system for lung tumor detection based on PET/CT images

The development of positron emission tomography / computed tomography (PET / CT) has shown great success in improving the accuracy of clinic diagnosis of lung cancers. However, it is still difficult to differentiate between some normal tissues with high standard uptake value (such as inflammations) and lung tumors by general methods. The objective of this paper was to identify textural features useful in distinguishing tumor from normal tissue in PET/CT images. A computer aided diagnosis system is developed. Tumors and normal tissues were segmented by optical threshold method. Texture features were computed from every segmented regions of interest (ROI), and then analyzed according to classification of ROIs. Finally, the effectiveness of distinguishing tumor and normal tissue of every feature was compared and analyzed using distance and KNN classifier. The clinic use of the system has potential to improve in the accuracy of discriminating benign and malignant lesions.