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

Abstract

Chest Radiograph or Chest X-ray (CXR) is a common, fast, non-invasive, relatively cheap radiological examination method in medical sciences. CXRs can aid in diagnosing many lung ailments such as Pneumonia, Tuberculosis, Pneumoconiosis, COVID-19, and lung cancer. Apart from other radiological examinations, every year, 2 billion CXRs are performed worldwide. However, the availability of the workforce to handle this amount of workload in hospitals is cumbersome, particularly in developing and low-income nations. Recent advances in AI, particularly in computer vision, have drawn attention to solving challenging medical image analysis problems. Healthcare is one of the areas where AI/ML-based assistive screening/diagnostic aid can play a crucial part in social welfare. However, it faces multiple challenges, such as small sample space, data privacy, poor quality samples, adversarial attacks and most importantly, the model interpretability for reliability on machine intelligence. This paper provides a structured review of the CXR-based analysis for different tasks, lung diseases and, in particular, the challenges faced by AI/ML-based systems for diagnosis. Further, we provide an overview of existing datasets, evaluation metrics for different[][15mm][0mm]Q5 tasks and patents issued. We also present key challenges and open problems in this research domain.

Keywords: COVID-19; Pneumoconiosis; chest X-ray; interpretable deep learning; pneumonia; trusted AI; tuberculosis.

Figure 1

Showcasing the chest-X rays for three projections. (A) AP view, (B) PA view, and (C) Lateral View.

Figure 2

Showcasing the transition across different tasks in CXR-based analysis for a given input image.

Figure 3

Showcasing research problems which have been studied in the literature.

Figure 4

Illustrating the schematic structure of the paper.

Figure 5

Showcases the examples of outputs obtained after tasks such as pre-processing and classification. (A) shows output of contrast enhancement. (B) shows output of the segmentation task and (C) shows the classification pipeline.

Figure 6

Showcasing the synthetically generated chest-X ray images. For a given normal image (A), the proposed approach by Tang et al. (2019b) generates the abnormal images (B–G) is predicted segmentation mask for same input image and results in mask-image pairs from (B–G). Figure is adapted from Tang et al. (2019b).

Figure 7

Showcasing the chest-X rays affected with different lung disorders. (A) Normal, (B) Pneumoconoisis, (C) TB, (D) Pneumonia, (E) Bronchitis, (F) COPD, (G) Fibrosis, and (H) COVID-19.

Figure 8

Showcases the sample examples of the CXRs from different datasets. The samples belong to Shenzhen (A, B), Montgomery (C, D), JSRT (E, F), Chestxray14 (G, H), VinDr-CXR (I, J), CheXpert (K, L), RSNA Pneumonia (M, N), Covid-CXR (O, P), PedPneumonia (Q, R), and MIMIC-CXR (S, T). The samples across different datasets highlight a wide variety in terms of quality, contrast, brightness and original image size.