CoISAE-PARCNN: Combined Improved Stacked Auto Encoder and Enhanced Pre-Activation Residual Convolutional Neural Network for the Pulmonary Nodule Detection in Lung CT and X-ray Images with robust image enhancement and segmentation techniques
Main Article Content
Abstract
Lung Cancer is one among deadly disease since it is aggressive in nature as well as deferred detections at advanced stages. The survival rate due to lung cancer mainly relies on prompt detection which is regarded as major challenge. Former approaches such as iW-Net based segmentation and Enhanced Inception-Residual Convolutional Neural Network (EIRCNN) based lung nodule detection are utilized for lung cancer detection. Though traditional iW-Net architecture outperforms in an improved way for CT medical images with challenging datasets through extensive experimentations, certain aspects are lagging pertaining to architecture. Yet further examination is required for optimal features selection in Lung nodule detection. To mitigate these issues, combination of Improved Stacked Autoencoder and Enhanced Pre-Activation Residual Convolutional Neural Network (EPARCNN) is greatly utilized for pulmonary nodule detection. Also robust image enhancement in lung CT and X-ray is achieved by suggesting Fuzzy Normalized Gamma-Corrected Contrast-Limited Adaptive Histogram Equalization (FNGCCLAHE) with Nonsubsampled Contourlet Transform (NSCT) and segmentation through Sub-Intensity range-based Pulse-Coupled Neural Network (SIPCNN). LIDC-IDRI and X–ray image datasets are utilized for assessing suggested CoISAE-PARCNN system with respect to factors such as precision, recall, f-measure, Peak Signal-o-Noise Ratio (PSNR), time, error rate and accuracy. It is thereby validated through experimental outcomes that the proposed system outperforms well in contrary to various prevailing approaches like LOG, Fast-RCNN, EWRCNN and EIRCNN.
Article Details
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.