Table of Contents
Coronavirus disease 2019 (COVID-19) has significantly affected the healthcare systems with tremendous impact on populations around the world. It has become need of the hour to develop fast and effective screening tools that could allow timely identification of patients infected with COVID-19 and help in achieving isolation and treatment.
Currently, the primary method of screening for COVID-19 is reverse transcription polymerase chain reaction (RT-PCR) testing that assesses SARS-CoV-2 ribonucleic acid (RNA) in sputum samples (from the upper respiratory tract). Although, RT-PCR testing is highly specific for COVID-19, but depending on the sampling method and time since onset of symptoms, the sensitivity varies.
Challenges and limitations of existing techniques
Literature shows that clinical signs and symptoms indicative of COVID-19 infection can be examined in chest CT images, with patchy shadows, crazy-paving pattern, ground-glass opacities and consolidation, mainly with bilateral involvement. Furthermore, it is observed that despite a negative RT-PCR test, patients may still present with abnormalities in chest due to COVID-19 infection.
However, it becomes quite challenging for the radiologist to differentiate the imaging abnormalities specific to COVID-19 infection and those caused due to other medical problems. Therefore, the skills and observations of radiologist to correctly analyze the visual differences may vary considerably.
Importance of deep learning model
Correct interpretation of the subtle differences based on radiographic examination is a demanding process for physicians. To remove this bottleneck, successful implementation of artificial intelligence (AI) based systems has been conducted to examine COVID- 19 from radiography images. Moreover, AI systems are fast and more efficient than the traditional examination where manual approach involves examination by a radiologist. A deep learning model will facilitate early detection of COVID-19 infection.
There are limited datasets that are publicly available owing to the recent outbreak of COVID-19. Researchers have no well-defined classification and segmentation method for this new disease. CT images in conjunction with deep learning models are being used by most of the investigators to differentiate COVID-19 patients from typical pneumonia patients. As CT scans are more effective than CT images, it is hypothesized that deep-learning techniques performed on chest CT scans will help physicians to properly assess and classify COVID-19, pneumonia, and healthy patients.
Advantages of AI in determining COVID severity
Deep learning models based on chest CT imaging have been proposed by several studies, making it easy to distinguish non-COVID-19 cases (including normal and abnormal cases) from COVID-19 cases. Furthermore, the existing proposed systems are able to identify non-COVID-19 cases as normal, non-pneumonia cases and non-COVID-19 pneumonia (such as community acquired pneumonia (CAP), bacterial pneumonia, viral pneumonia).
A recent study integrated chest CT findings with different factors including clinical symptoms, exposure history and laboratory testing by using artificial intelligence (AI) algorithms to quickly detect patients who are positive for COVID-19.
Conclusion
Deep learning model has proved beneficial as it conducts automated measurements (wall thickness, lobe segmentation etc), improves image visualization and prevent chances of human error. Beside clinical procedures and treatments, successful adoption of AI will show promising results in healthcare organizations. It will help in designing different tools based upon machine learning (ML) algorithms and enhance the decision-making processes. Deep learning models will help to segregate patients depending on disease severities and prepare structured report, based on individualized assessment of the extent of the lesions.
