Zhonghua Yi Xue Za Zhi
December 2009
Zhonghua Wai Ke Za Zhi
October 2007
Objective: To evaluate the efficacy of the digital cytopathological lung cancer diagnosing system (DCLCDS) utilizing the latest computer technologies (including reinforcement learning, image segmentation and classifier) and the cytopathological knowledge on lung cancer cells.
Methods: Separate the overlapped lung cancer cells in a slice image applying the improved deBoor-Cox B-Spline algorithm; Segment cell regions in a slice image using an image segmentation algorithm based on reinforcement learning; Ensemble different classifiers, including Decision Tree classifier, Support Vector Machine (SVM) classifier and Bayesian classifier, to achieve an accurate result of cytopathological lung cancer diagnosis.
Results: The accurate diagnosis rate for lung cancer identification of 224 images of small lung lesions aspiration biopsy from 120 cases randomly selected was 92.