FISSURE DETECTION IN LUNG CT IMAGES USING BACK PROPAGATION NEURAL NETWORKS
Fissure segmentation and detection of fissures in lung CT images is essential in the clinical practice for the interpretation and diagnosis of diseases present in the lungs. A novel supervised learning approach using multilayer back propagation neural networks is employed to segment the fissures. The pixels present on the fissures are classified based on the feature set calculation. An optimized feature subset selection is done to enhance the performance. The classifier is trained with the known input to achieve the targeted output. The ground truth for the classification is provided by an expert observer. The method overcomes the necessity of manual intervention required in training the classifier used in the previous conventional methods. The back propagation neural network adjusts the weights and minimizes the mean square error to overcome the tedious process of selecting the training data to train the classifier. The method was investigated with 365 CT images from 15 CT examinations. The performance index of the proposed method is measured to be good with an area 0.98 under ROC curve.
DEVAKI K , AKILA DEVI R AND VIDHUPRIYA P