Based on a chest CT alone, an artificial intelligence (AI) algorithm can predict clinical outcomes for COVID-19 patients such as the need for admission to the ICU, according to recent research.
COVID-19 diagnosis is typically made through a PCR test, a nasal swab that detects bits of viral genetic material. While chest CT is generally not used as a first-line tool to diagnose or screen for COVID-19, it can help assess the seriousness of the disease and determine a plan of care for patients.
Xu
While AI-based deep learning (DL) has the potential to aid in automated and standardized evaluation of CT scans, RSNA 2020 researchers sought to determine the potential of the DL method to predict clinical outcomes of patients with COVID-19 based on chest CT.
In research presented at RSNA 2020, Ziyue Xu, PhD, and a multinational team of researchers recently investigated this potential in a study that drew upon thousands of images from China, Japan, Italy and the United States.
"While there are many efforts using AI to detect COVID-19, there are still critical challenges due to data diversity issues," said Dr. Xu, senior scientist with NVIDIA Corporation, a multinational technology company based in Santa Clara, CA. "That is why we chose to collect data from diverse sources."
In the study, expert radiologists carefully annotated the images and used them to train the AI algorithm to identify subtle features that might help predict outcomes for COVID-19 patients.
The researchers correlated 632 chest CT scans in patients diagnosed with COVID-19 with the clinical outcome. Of the 632 patients, 69 were admitted to the ICU.
Algorithm Highly Effective in Predicting Outcomes
The model achieved an overall accuracy of 92%. It had a sensitivity of 73% and a specificity 93.9%, despite an imbalanced dataset weighted towards no ICU admission, the authors said.
While the positive predictive value for ICU admission was only 53%, the negative predictive value was 97%, meaning the algorithm was highly effective at identifying patients who would not need admission to the ICU.
"Based on chest CT alone, AI-based deep learning algorithms can reasonably predict clinical outcomes such as ICU admission in patients with COVID-19 who underwent CT and PCR on the day of admission," Dr. Xu said.
Dr. Xu cautioned that the current study is preliminary and has limitations, the technology has tremendous potential to speed care to patients suffering from the infectious disease that has killed more than one million people worldwide.
"The model is feasible with reasonable accuracy and specificity of prediction," Dr. Xu said. "Such a model might alert the clinician to the enhanced potential of ICU admission, especially if combined with other clinical features."
By providing a better understanding of the appearance of COVID-19 in chest CT, the model could also serve as a tool for mitigating outbreaks, Dr. Xu added.
For More Information
View the RSNA 2020 session Artificial Intelligence Deep Learning with Chest CT to Predict Outcomes Such as ICU Admission in COVID-19: A Multinational Study — SSCH03 at RSNA2020.RSNA.org.
