COVID-19 identification in chest X-ray images on flat and hierarchical classification scenarios

Abstract

Background and objective: The COVID-19 can cause severe pneumonia and is estimated to have a high impact on the healthcare system. Early diagnosis is crucial for correct treatment in order to possibly reduce the stress in the healthcare system. The standard image diagnosis tests for pneumonia are chest X-ray (CXR) and computed tomography (CT) scan. Although CT scan is the gold standard, CXR are still useful because it is cheaper, faster and more widespread. This study aims to identify pneumonia caused by COVID-19 from other types and also healthy lungs using only CXR images.

Methods: In order to achieve the objectives, we have proposed a classification schema considering the following perspectives: i) a multi-class classification; ii) hierarchical classification, since pneumonia can be structured as a hierarchy. Given the natural data imbalance in this domain, we also proposed the use of resampling algorithms in the schema in order to re-balance the classes distribution. We observed that, texture is one of the main visual attributes of CXR images, our classification schema extract features using some well-known texture descriptors and also using a pre-trained CNN model. We also explored early and late fusion techniques in the schema in order to leverage the strength of multiple texture descriptors and base classifiers at once. To evaluate the approach, we composed a database, named RYDLS-20, containing CXR images of pneumonia caused by different pathogens as well as CXR images of healthy lungs. The classes distribution follows a real-world scenario in which some pathogens are more common than others.

Results: The proposed approach tested in RYDLS-20 achieved a macro-avg F1-Score of 0.65 using a multi-class approach and a F1-Score of 0.89 for the COVID-19 identification in the hierarchical classification scenario.

Conclusions: As far as we know, the top identification rate obtained in this paper is the best nominal rate obtained for COVID-19 identification in an unbalanced environment with more than three classes. We must also highlight the novel proposed hierarchical classification approach for this task, which considers the types of pneumonia caused by the different pathogens and lead us to the best COVID-19 recognition rate obtained here.

Keywords: COVID-19; Chest X-ray; Medical image analysis; Pneumonia; Texture.

Graphical abstract

Fig. 1

The hierarchical class structure of pneumonia caused by micro-organisms.

Fig. 2

Different classes distribution in a binary labeled dataset.

Fig. 3

Example of datasets before and after applying the resampling techniques.

Fig. 4

The proposed classification schema for the COVID-19 identification in CXR images.

Fig. 5

Example of combinations with late fusion strategies using the sum, product and voting strategies. The example dataset has M samples and L labels.

Fig. 6

RYDLS-20 image samples.

Fig. 7

F1-Score results per label in the best case scenario for multi-class context.

Fig. 8

Confusion Matrix in the best case scenario for the multi-class experiments.

Fig. 9

F1-Score results per label in the best case on the hierarchical scenario.

Fig. 10

Confusion Matrix in the best case scenario for the hierarchical experiments.

Fig. 11

Best F1-Score results on multi-class and hierarchical scenarios for COVID-19 Identification.

Fig. 12

Best macro-avg F1-Score results on multi-class and hierarchical scenarios.

Fig. 13

Examples of samples with “COVID-19” label that were predicted as “Normal”.

Fig. 14

Different examples of CXR with “normal” lungs.