Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

Artificial Intelligence May Aid in the Discrimination of Radiation Pneumonitis from COVID-19-Associated Interstitial Pneumonia

A deep-learning algorithm can help to discriminate radiation pneumonitis from COVID-19 pneumonia
24 Mar 2021
COVID-19 and Cancer;  Cancer Intelligence (eHealth, Telehealth Technology, BIG Data)
Thoracic Malignancies

An artificial intelligence (AI) algorithm was able to classify most patients with radiation induced pneumonitis into a category of low risk for Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) to assist in differentiating between the syndromes, according to findings presented at the European Lung Cancer Virtual Congress 2021 (25-27 March).

Sara Ramella of the Radiation Oncology, Campus Bio-Medico University in Rome, Italy underscored the importance of distinguishing between SARS-CoV-2-associated severe interstitial pneumonia and radiation pneumonitis, which demonstrate similar and overlapping clinical features. Prof. Ramella and colleagues conducted this study to determine whether a deep learning algorithm may aid in the determination of radiation pneumonitis from COVID-19 pneumonia.

The study analysed the data and computed tomography (CT) images of 34 patients with COVID-19 pneumonia and 36 with radiation pneumonitis. The InferReadTM CT Lung (COVID-19) ®, an Artificial Intelligence algorithm based on a novel deep convolutional neural network structure, was used to analyse the patients’ CT images. Cut-off for the estimated risk probability of COVID-19 was set at levels higher than 30% and categorised as COVID-19 high risk, according to a recent publication which determined the percentage of COVID-19 confirmed patients above this cut-off value was higher than 95%. Using the cut-off set in the same publication, values of estimated risk probability below 30% were classified as COVID-19 low risk.

Artificial-Intelligence-May-Aid-in-the-Discrimination-of-Radiation-Pneumonitis-from-COVID-19-Associated-Interstitial-Pneumonia-Figure-1

InferReadTM CT Lung (COVID-19) system interface example; comparison between a COVID-19 pneumonia positive patient (A) and a radiation pneumonitis-affected patient (B).

© Sara Ramella.

Statistical analyses included the Mann Whitney U test using a significance threshold of p < 0.05, and receiver operating characteristic (ROC) curve with fitting performed by using the maximum likelihood fit of a binormal model.

The algorithm aided the differential diagnosis of radiation- and COVID-19-associated pneumonia

According to the algorithm, 66.7% of patients with radiation pneumonitis were classified as COVID-19 low risk and all patients with radiation pneumonitis that were classified as COVID-19 high risk had ≥grade 3 disease.

The algorithm showed good accuracy in the detection of radiation pneumonia apart from COVID-19 pneumonia with 97% sensitivity and 2% specificity (area under the curve [AUC] 0.72). Accuracy was increased with the application of a 30% cut-off to 76% sensitivity and 63% specificity (AUC 0.84). The investigators also found that total lung volume (p = 0.001), the left lower lobe (p < 0.001), and the right lower lobe (p < 0.001) involvement were increased in the COVID-19 group compared to patients with radiation pneumonitis.

Artificial-Intelligence-May-Aid-in-the-Discrimination-of-Radiation-Pneumonitis-from-COVID-19-Associated-Interstitial-Pneumonia-Figure-2

Receiving operating curves (ROC) of the diagnostic performance of the artificial intelligence prediction risk of COVID-19 pneumonia. Each plot shows the ROC obtained on the testing after including the pairs: COVID-19 and pneumonia-free patients (A), RP and COVID-19 (B), COVID-19 and non-COVID-19 patients (C), RP and COVID-19 with 30% threshold (D), COVID-19 and RP total lung volume involvement (E), COVID-19 and RP RUL involvement (F), COVID-19 and RP RLL involvement (G), and COVID-19 and RP LLL involvement (H). Gray lines plot 95% confidence intervals.

RP - Radiation Pneumonitis, RUL - Right Upper Lobe, ML - Middle Lobe, RLL - Right Lower Lobe, LUL - Left Upper Lobe, LLL - Left Lower Lobe, AUC = area under ROC, Std. Error = Standard Error.

© Sara Ramella.

Conclusions

According to the authors, a deep-learning algorithm can help to discriminate radiation pneumonitis from COVID-19 pneumonia, and was able to classifying most radiation pneumonitis as low-risk COVID19. They advise that, in cases where patients treated with radiation therapy are classified as high risk, dosimetric factors should be taken into account.

They further recommend that, in cases where patients pretreated with radiotherapy and presenting with diffuse pneumonitis are classified by AI as COVID-19 high risk, a combination of dosimetric factors may help to identify radiation pneumonitis, such as an increase in positive predictive value from 60% to 99.8%.

No external funding was disclosed.

Reference

42P – Ramella S, Quattrocchi CC, Ippolito E, et al. Radiation induced pneumonitis in the era of COVID-19 pandemic: Artificial intelligence for differential diagnosis. European Lung Cancer Virtual Congress 2021 (25-27 March).

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.