Hybrid Deep Learning based Lung Disease Detection and Classification

Authors

  • N. Malligeswari, Dr. G. Kavya, M. Papitha, G. S. Joseph

Abstract

Early precise detection of lung nodule is overwhelming of time and susceptible to error factor of the radiologist analysis work. Recent lung nodule detection are based on CNN and Faster RCNN results good accuracy and superior performance in classification. However, this is an object detection algorithm means find where the objects are present. Apart from object detection Mask RCNN is an extension of faster RCNN implementing segmentation on image means separating the pixels that belongs to particular object .Image segmentation work with the help of Mask RCNN, We can perform object detection and also specifically locate the position of cancer tumor in lung. We proposed a 3D Mask RCNN for Simultaneous detection and Segmentation of lung nodule probing more number of training and testing data to reduce the false positive reduction and achieve higher accuracy and sensitivity. For further advancing the performance of our work we investigated more than 2000 ground truth nodules from publically available LIDC/IDRI dataset advantageous to boost our Mask RCNN detection. Experiment results shows that the proposed network succeeds accuracy of 96.8%., sensitivity of 94.8% and specificity of 97.2%. After evaluation and investigation the results of segmentation our proposed method outperformed compared to other literature.

Downloads

Published

2022-08-09