New AI Network Predicts Covid-19 Outcomes

Researchers at Massachusetts General Hospital (MGH( have developed an “unsupervised” AI network (one trained without the use of annotated CT scans) to predict the outcomes of Covid-19 patients.

“In this study, we developed a weakly unsupervised conditional GAN [generative adversarial netowk], called pix2surv, which enables the estimation of the distribution of the survival time directly from the chest CT images of patients,” wrote lead author Tomoki Uemura and his team. Doing so enabled them to overcome the limitations of previous predictive models because the model allows them to train the network using only unannotated, 2D CT slices and the associated survival times of the patients.

Highlights:

  • Prediction of COVID-19 progression and mortality based on chest CT images.
  • Outperforms existing laboratory and image-based prognostic predictors for COVID-19.
  • Enables stratification of the patients into low- and high-risk groups by a wider margin than those of the other predictors.

The authors conclude that “pix2surv is a promising approach for performing image-based prognostic prediction for the management of patients.”

Source:

Tomoki Uemura, Janne J. Näppi, Chinatsu Watari, Toru Hironaka, Tohru Kamiya, Hiroyuki Yoshida, “Weakly unsupervised conditional generative adversarial network for image-based prognostic prediction for COVID-19 patients based on chest CT, Medical Image Analysis, Volume 73, 2021, https://doi.org/10.1016/j.media.2021.102159. Available online July 11, 2021, at https://www.sciencedirect.com/science/article/pii/S136184152100205X Accessed August 23, 2021

An example of the predicted overall survival time (mortality) of a 67-year old male who expired 10 days (red dotted line on the plot on the right) after the chest CT examination. The image on the left shows a representative example of the CT images. The plot on the right shows the predicted survival times (circles) by pix2surv and the image-based reference predictors, with 95% confidence interval bars superimposed on the boxplots that represent the bootstrap results. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Image and caption used under terms of the Creative Commons CC-BY license. The citation is in the main article. The picture was resampled to 144ppi; the caption had the figure number removed.