Lung Cancer Classification using Gray-Level Co-Occurrence Matrix Feature Extraction and Forward Selection Feature Selection based on the K-Nearest Neighbor Algorithm

<p class="Abstract">In diagnosing lung cancer, the <em>medical imaging team </em>manually identifies CT-scan<em> images of the lungs. </em>This identification process makes it difficult for the <em>medical imaging team </em>to differentiate between lung cancer and normal images. This is because there is <em>noise </em>in the image, which reduces the image quality, so <em>image processing must reduce the noise</em>. This study used median and Gaussian filters, Otsu thresholding segmentation, GLCM feature extraction, forward selection, and k-nearest Neighbor classification. The research results show that of the 22 statistical features extracted, only 16 were selected for characterizing image classification. The image datasets used are 900 image data sets for <em>program training </em>and 100 image data sets for <em>program testing</em>. With a dataset of 100 image data sets, the level of diagnostic accuracy without <em>forward selection </em>(22 GLCM features) was 81.67%, while the diagnostic accuracy using <em>forward selection </em>(16 GLCM features) was 93.22% with a sensitivity of 92.25% and specificity is 94.46%.</p>