Deteksi Nodul Paru pada Citra CT dengan Klasifikasi Pseudo Nearest Neigbour Rule

Abstract

This research aims to obtain the classification performance of the Pseudo Nearest Neighbor Rule (PNNR) algorithm in detecting lung nodules in CT scan images. The PNNR classification algorithm is used to reduce the influence of noise or outliers in the classification process so that false positives (prediction of an object that is not a nodule as a nodule) can be reduced. The data set used is 200 patient data obtained from the public data of The Lung Image Database Consortium and Infectious Disease Research Institute (LIDC/IDRI) where 4 fold Cross Validation will be carried out. The preprocessing stage is carried out by segmenting the otsu image, where from the segmentation results the two largest blobs are then searched for to determine the area of ​​the lung to be analyzed. Next, the feature extraction process from the candidate nodules (white pixels / foreground) is obtained from the Otsu segmentation process again. The results of this second segmentation contain information from the candidate nodules to then calculate the value of the shape features of the candidate nodules such as area, eccentricity, equivalent diameter, major axis length, minor axis length and perimeter which produces feature set values ​​as the basis for training data and data test for the classification process in PNNR The results of the classification proposed in this research, namely using the PNNR classification method, obtained an Accuracy value of , which is included in the excellent classification level or the Accuracy level is very good but with a lower level of sensitivity or recognition of true positives, namely . In further research, classification optimization can be carried out by selecting the feature set usedThis research aims to obtain the classification performance of the Pseudo Nearest Neighbor Rule (PNNR) algorithm in detecting lung nodules in CT scan images. The PNNR classification algorithm is used to reduce the influence of noise or outliers in the classification process so that false positives (prediction of an object that is not a nodule as a nodule) can be reduced. The data set used is 200 patient data obtained from the public data of The Lung Image Database Consortium and Infectious Disease Research Institute (LIDC/IDRI) where 4 fold Cross Validation will be carried out. The preprocessing stage is carried out by segmenting the otsu image, where from the segmentation results the two largest blobs are then searched for to determine the area of ​​the lung to be analyzed. Next, the feature extraction process from the candidate nodules (white pixels / foreground) is obtained from the Otsu segmentation process again. The results of this second segmentation contain information from the candidate nodules to then calculate the value of the shape features of the candidate nodules such as area, eccentricity, equivalent diameter, major axis length, minor axis length and perimeter which produces feature set values ​​as the basis for training data and data test for the classification process in PNNR The results of the classification proposed in this research, namely using the PNNR classification method, obtained an Accuracy value of , which is included in the excellent classification level or the Accuracy level is very good but with a lower level of sensitivity or recognition of true positives, namely . In further research, classification optimization can be carried out by selecting the feature set used