A fuzzy-enhanced deep learning approach for early detection of Covid-19 pneumonia from portable chest X-ray images

Abstract

The Covid-19 pandemic is the defining global health crisis of our time. Chest X-Rays (CXR) have been an important imaging modality for assisting in the diagnosis and management of hospitalised Covid-19 patients. However, their interpretation is time intensive for radiologists. Accurate computer aided systems can facilitate early diagnosis of Covid-19 and effective triaging. In this paper, we propose a fuzzy logic based deep learning (DL) approach to differentiate between CXR images of patients with Covid-19 pneumonia and with interstitial pneumonias not related to Covid-19. The developed model here, referred to as CovNNet, is used to extract some relevant features from CXR images, combined with fuzzy images generated by a fuzzy edge detection algorithm. Experimental results show that using a combination of CXR and fuzzy features, within a deep learning approach by developing a deep network inputed to a Multilayer Perceptron (MLP), results in a higher classification performance (accuracy rate up to 81%), compared to benchmark deep learning approaches. The approach has been validated through additional datasets which are continously generated due to the spread of the virus and would help triage patients in acute settings. A permutation analysis is carried out, and a simple occlusion methodology for explaining decisions is also proposed. The proposed pipeline can be easily embedded into present clinical decision support systems.

Keywords: Chest X-ray; Convolutional Neural Network; Covid-19; Fuzzy logic; Portable systems; explainable Artificial Intelligence.

Fig. 1

Examples of original (a) and pre-processed (b) CXR images.

Fig. 2

Examples of pre-processed CXR images (a) and fuzzy CXR images (b), obtained by applying the proposed fuzzy edge detection procedure.

Fig. 4

Fuzzy-enhanced CNN classification system (approach 1): CXR and fuzzy images of the same patient are stratified in a volume of data $v_1$ × $v_2$ × $v_3$ = 800 × 900 × 2 and used as input to the proposed CovNNet followed by a standard MLP for performing the 2-way classification task: Covid-19 vs. No-Covid-19.

Fig. 5

Fuzzy-enhanced CNN classification system (approach 2): CXR and fuzzy images of the same patient are used as input to two CovNNet here employed to automatically extracts the most relevant CXR-features from CXR images and fuzzy-features from fuzzy images. Such features are concatenated and used as input to a standard MLP for performing the 2-way classification task: Covid-19 vs. No-Covid-19.

Fig. 3

Lay-out of the Covid-19 vs. No-Covid-19 pneumonia classification system based on a CNN approach (CovNNet)..

Fig. 6

Saliency maps obtained for sample images that were correctly classified by the CNN. Each saliency map is overlapped to the corresponding RX image. The coloration of the pixels in the saliency map ranges from blue (low relevance) to red (high relevance). The true class (Covid-19/No-Covid-19) can be read in the title displayed on top of every image.

Fig. 7

Distribution of the FD values for Covid-19 (red) and No-Covid-19 (blue) CXR images. Dashed lines represent the average FD of Covid-19 and No-Covid-19 class.

Fig. 8

Feature maps learned by the three convolutional layers of CovNNet on a Covid-19 (a) and No-Covid-19 CXR image (b). Note that the convolutional layers generate four, eight and sixteen feature maps sized 400 × 450, 100 × 112 and 25 × 28, respectively. The learning procedure seems to assign a highest resolution to the feature maps generated from Covid-19 images. Some feature maps are evidently devoted to detect the zones where lung hilum, thickening of the lung texture, pulmonary fibrosis are present, some others simply highlight diffuse opacities bilaterally.