Delaunay triangulation based intelligent system for the diagnosis of covid from the low radiation CXR images

Abstract

Covid-19 is a viral infection that causes a profound impact on the lives of the World population. It is a global pandemic spreading across the world in a faster way. It made a global impact on the health, economy, and education system in all the countries. As it is a rapidly spreading disease, prevention demands a fast and accurate diagnosis system. In a highly densely populated country, the demand for fast and affordable early diagnosis is required to reduce the disaster. Within this diagnosis time, the infection spreads rapidly and worsens the infected person's status. To provide a faster and more affordable early diagnosis of covid, posterior-anterior chest radiographs (CXR) are used. Diagnosis of covid from CXR is challenging due to the images' interclass similarity and intraclass variation. This study proposes a deep learning-based robust early diagnosis method for covid. To balance the intraclass variation and interclass similarity in CXR images, the deep fused Delaunay triangulation (DT) is proposed as the CXR has low radiation and unbalanced quality images. The deep features are to be extracted to increase the robustness of the diagnosis method. Without segmentation, the proposed DT algorithm achieves the accurate visualization of the suspicious region in the CXR. The proposed model is trained and tested by the largest benchmark covid-19 radiology dataset with 3616 covid CXR images and 3500 standard CXR images. The performance of the proposed system is analyzed in terms of accuracy, sensitivity, specificity, and AUC. The proposed system yields the highest validation accuracy.

Keywords: Covid; Deep neural networks; Delaunay triangulation.

Fig. 1

Proposed dtXpert framework

Fig. 2

Intra-class variation of TB CXR images

Fig. 3

Reference points for CXR

Fig. 4

Interpolated Point P from P₁, P₂ and P₃

Fig. 5

Mapping of points P₁, P₂, and P₃ from the (x, y) plane into the (u, v) plane

Fig. 6

Overlay image based on DT Interpolated points

Fig. 7

Deep fused CXR image

Fig. 8

Intra-class variation of CXR images after DT

Fig. 9

Correlation coefficient of DTI covid dataset vs. original covid dataset

Fig. 10

Correlation coefficient of Covid vs. Normal images

Fig. 11

Interclass similarity and intraclass variation analysis

Fig. 12

Performance analysis of DTI with SqeenzeNet

Fig. 13

Performance analysis of DTI with Inception-ResNet V2

Fig. 14

Training performance analysis of DtXpert