Pneumonia Detection from Chest X-ray Images Based on Convolutional Neural Network

Abstract

Pneumonia has caused significant deaths worldwide, and it is a challenging task to detect many lung diseases such as like atelectasis, cardiomegaly, lung cancer, etc., often due to limited professional radiologists in hospital settings. In this paper, we develop a straightforward VGG-based model architecture with fewer layers. In addition, to tackle the inadequate contrast of chest X-ray images, which brings about ambiguous diagnosis, the Dynamic Histogram Enhancement technique is used to pre-process the images. The parameters of our model are reduced by 97.51% compared to VGG-16, 85.86% compared to Res-50, 83.94% compared to Xception, 51.92% compared to DenseNet121, but increased MobileNet by 4%. However, the proposed model’s performance (accuracy: 96.068%, AUC: 0.99107 with a 95% confidence interval of [0.984, 0.996], precision: 94.408%, recall: 90.823%, F1 score: 92.851%) is superior to the models mentioned above (VGG-16: accuracy, 94.359%, AUC: 0.98928; Res-50: accuracy, 92.821%, AUC, 0.98780; Xception: accuracy, 96.068%, AUC, 0.99623; DenseNet121: accuracy, 87.350%, AUC, 0.99347; MobileNet: accuracy, 95.473%, AUC, 0.99531). The original Pneumonia Classification Dataset in Kaggle is split into three sub-sets, training, validation and test sets randomly at ratios of 70%, 10% and 20%. The model’s performance in pneumonia detection shows that the proposed VGG-based model could effectively classify normal and abnormal X-rays in practice, hence reducing the burden of radiologists.

AI-based radiodiagnosis using chest X-rays: A review

“…Sherrier and Johnson ( 1987 ) used a region-based histogram equalization technique to improve the image quality of CXR locally and finally obtain an enhanced image. Zhang D. et al ( 2021 ) used the dynamic histogram enhancement technique (Abin et al, 2022 ) used different image enhancement techniques such as Brightness Preserving Bi Histogram (BBHE) (Zadbuke, 2012 ), Equal Area Dualistic Sub-Image Histogram Equalization (DSIHE) (Yao et al, 2016 ), Recursive Mean Separate Histogram Equalization (RMSHE) (Chen and Ramli, 2003 ) followed by a Particle swarm optimization (PSO) (Settles, 2005 ) for further enhancing the CXRs for detecting pneumonia. Soleymanpour et al ( 2011 ) used adaptive contrast equalization for enhancement, morphological operation-based region growing to find lung contour for lung segmentation followed by oriental spatial Gabor filter (Gabor, 1946 ) for rib suppression.…”

Investigating the Effectiveness of Interpretable Cost-Sensitive Neural Network for Pneumonia Detection Contemplating Data Imbalance

“…3 An early diagnosis of pneumonia can be helped by a cost-sensitive neural network, and pneumococcal immunizations can avoid over 400,000 child deaths each year. 4 Early diagnosis of pneumonia is essential for full recovery since it enables early and effective medical intervention. The more quickly the infection is identified and diagnosed, may be treated to stop its spread and reduce its severity.…”

A systematic survey on deep learning techniques for chest disease detection using chest radiographs

“…The system acquired 98.81% and 86.85 % accuracy, and 98.80% and 87.02% sensitivity on Kermany and RSNA datasets respectively. Zhang et al, (2021) proposed a VGG-based model with fewer layers and dynamic histogram equalization approach has been applied for the pre-processing purpose that is helpful for pneumonia identification. Manickam et al, (2021) presented a system for detecting pneumonia infection on CXRs that uses U-Net architecture-based segmentation and classifies pneumonia using pre-trained ImageNet dataset models.…”