Deep volcanic residual U-Net for nodal metastasis (Nmet) identification from lung cancer

Abstract

Lymph node metastasis detections are more clinically significant task associated with the presence and reappearance of lung cancer. The development of the computer-assisted diagnostic approach has greatly supported the diagnosis of human disorders in the field of medicine including lung cancer. Lung cancer treatment is possible if it is detected at the initial stage. Radiologists have great difficulty identifying and categorizing lung cancers in the initial phase. So, several methods were used to predict the lung cancer but does not provide accurate solutions with increased error rate. To overcome these issues, a Deep Volcanic Residual U-Net (DVR U-Net) for nodal metastasis is proposed in this manuscript which identifies the LC accurately in the early stage. Initially, the input images are taken from two datasets. After that, these input data are pre-processed using Anisotropic Diffusion Filter with a Fuzzy based Contrast-Limited Adaptive Histogram Equalization (ADFFCLAHE) method. Then the pre-processed images are given to the DVR U-Net to segment and extract the volume of interest for estimating the nodal stage of each volume of interest. Finally, DVR U-Net effectively detects and classifies the N + (nodal metastasis) or N- (non-nodal metastasis). The introduced method attains 99.9% higher accuracy as compared with the existing methods. Also, the statistical analysis of the Shapiro-Wilk test, Friedman test and Wilcoxon Signed-Rank test are executed to prove the statistical effectiveness of the implemented method.

Keywords: Deep Residual U-Net; Lymph node metastasis identification; Node metastasis; Non-Node metastasis; Volcano eruption algorithm.

Fig. 1

Overall flow diagram of the proposed methodology

Fig. 2

Flowchart of DRU-NET

Fig. 3

Output results of N + (Nmet) using LIDC/IDRI dataset

Fig. 4

Output results of N− (Non-Nmet)) using LIDC/IDRI dataset

Fig. 5

Output results of N + (Nmet) using dual-energy CT image dataset

Fig. 6

Output results of N− (Non-Nmet) using dual-energy CT image dataset

Fig. 7

a, b, c, d: Performance analysis of similarity score and dice score of LIDC/IDRI and dual-energy CT image dataset

Fig. 8

a, b, c, d: Performance analysis of error rate and Hausdroff distance (HD) of LIDC/IDRI and dual-energy CT image dataset