Improved Classification and Accurate Identification of Lung Cancer stages in Computed Tomography images using modified Support Vector Machine Classifier
Automatic Lung cancer detection is a decisive demanding mission for researchers because of the noise signals getting included in creative signals amid the image acquisition process which may corrupt the quality of the tumor images there by reducing the image to a desecrated presentation. The Crucial parameters of the lung nodules are the size and shape of the nodules in the tumor diagnosis. In this proposed model, the pre-processing stage is enhanced by using the Median and Gaussian filter in place of the Gabor filter. The segmentation process of the preprocessed images is carried out using the region growing algorithm followed by the Otsu- thresholding method. The advantage of this proposed model is the better perceptibility of the cancer nodules. The image features like Centroid, Diameter and Pixel mean intensity of the Lung nodules are extracted for the classification process. The existing model trims the process after the detection of cancer nodule, its feature extraction and calculation of accuracy; whether the detected nodule is cancerous or non-cancerous was not discussed. In this proposed algorithm, the process continues after the detection of the infected lung nodules by the segmentation procedures and also detects the extracted nodules as cancerous or non- cancerous also state the stage of the tumour. For the evaluation of the proposed algorithm, LIDC (Lung Image database Consortium) datasets and Cancer Imaging Archive (CIA) datasets. Therefore, in the proposed methodology, a further stage of cancer nodule identification using the extracted features is executed using Support Vector Machine Classifier. The extracted features after the classification process are used as training features and a trained model is generated for the classification of the detected nodule as a Cancerous or Non-cancerous Nodules. The proposed model detects the cancer with an accuracy of 97.93%, which is higher than the prevailing current model and the classifier has the prediction accuracy of 95.05% with minimum classification error of 0.039.
Keywords- Median Filter, Gabor Filter, Otsu's Thresholding, LIDC & CIA Datasets