Residual Partial Derivative Multilayer Perceptron Deep Learning Based Classifier for Lung Cancer Detection

Abstract

Lung cancer is a significant and frequent types of cancer globally both in patients and mortality rate. Computed Tomography (CT) test is efficient for early lung cancer detection due to potentiality in constructing lung images, generating nodules both in high resolution and tumor detection. But, early detection of lung cancer is obtained as difficult and time-consuming task. Also, detection of pulmonary nodules is a main method for early detection of lung cancer that can even heavily improve survival rate of lung cancer patients by reduced mortality rate to better level. Moreover, correct precision and incorrect detection of positive cancerous cell for lung nodules is both low and high. A lung cancer detection method using deep learning called, Residual Neural and Partial Derivative Multilayer Perceptron (RN-PDMP) is proposed. Initially the aid of lung cancer images acquired from IQ-OTH/NCCD Lung Cancer Dataset addresses class imbalance issue, therefore reducing the noise to a greater extent using Cross Entropy Focal Loss-based preprocessing model. Secondly preprocessed images as input, robust features are extracted by means of Census Transformed Residual Neural Network-based feature extraction algorithm. Finally, classification between normal, benign and malignant are made using Partial Derivative Multilayer Perceptron Deep Learning-based Classifier. The performance analysis of results describe to lung cancer detection model based on RN-PDMP integrating Forward Propagation-based Residual Learning has achieved a precision of 99% and recall of 97%. The results are comparatively better than existing methods in RN-PDMP method of lung cancer detection based on Partial Derivative-based back propagation is better recall for small samples and improves true positive pulmonary nodules.