<![CDATA[Purpose: This research aims to produce the best performance in identifying early-stage lung cancer class through CT-Scan image analysis using the decision tree classification method and to determine the results of the best classification performance from the variations carried out.Methods: Five steps in the CT-Scan image classification process for early-stage lung cancer class based on tumor density measurements. First, image data preparation where the image data used was 280 CT-Scan images with a pixel size of 607 x 607 and PNG format taken from the LIDC-IDRI database at https://www.cancerimagingarchive.net/ with a total of 1010 CT-Scan data scans. Second, the grayscaling stage converts the RGB image to a grayscale. Third, combining a high pass filter and Gaussian smoothing filter method is used to remove salt pepper noise and to smooth the image. Fourth, the feature extraction stage uses first and second-order statistics with 22 features used. The fifth is the classification stage using a decision tree, which is then validated using the k-fold method with k=10 so that all image data can be tested thoroughly.Result: The accuracy rate at the training stage was 90.51%, and at the testing stage was 89.99%. Stage I lung cancer detection program through CT-Scan imagery was successfully created with the highest PSNR value proven to optimize the accuracy level, precision, and recall in the testing phase results of 89.99%, 91.24%, and 89.64%.Novelty: Based on previous research searches, no one had used machine learning to classify early-stage lung cancer. Punithavathy et al. (2015) and Meliala (2021) stated that early detection of lung cancer can increase survival by 60%-70%. This research will produce a new method for determining early-stage lung cancer. ]]>
Copyrights © 2023
