Unsupervised AI Model Accurately Predicts COVID-19 Patient's Survival Based on Chest CT Exams
By MedImaging International staff writers
Posted on 08 Aug 2021
An "unsupervised" artificial intelligence (AI) model, or one trained without image annotations, can accurately predict the survival of COVID-19 patients on the basis of their chest computed tomography (CT) exams.
Researchers from Massachusetts General Hospital (Boston, MA, USA) have shown that the performance of their pix2surv algorithm based on CT images significantly outperformed those of existing laboratory tests and image-based visual and quantitative predictors in estimating the disease progression and mortality of COVID-19 patients. Thus, pix2surv offers a promising approach for performing image-based prognostic predictions.
Because of the rapid spread and wide range of the clinical manifestations of the coronavirus disease 2019 (COVID-19), fast and accurate estimation of the disease progression and mortality is vital for the management of the patients. Currently available image-based prognostic predictors for patients with COVID-19 are largely limited to semi-automated schemes with manually designed features and supervised learning, and the survival analysis is largely limited to logistic regression. To resolve this problem, the researchers developed a weakly unsupervised conditional generative adversarial network, called pix2surv, which can be trained to estimate the time-to-event information for survival analysis directly from the chest CT images of a patient.
pix2surv enables the estimation of the distribution of the survival time directly from the chest CT images of patients. The model avoids the technical limitations of the previous image-based COVID-19 predictors, because the use of a fully automated conditional GAN makes it possible to train a complete image-based end-to-end survival analysis model for producing the time-to-event distribution directly from input chest CT images without an explicit segmentation or feature extraction efforts. Also, because of the use of weakly unsupervised learning, the annotation effort is reduced to the pairing of input training CT images with the corresponding observed survival time of the patient.
In their study the researchers showed that the prognostic performance of pix2surv based on chest CT images compares favorably with those of currently available laboratory tests and existing image-based visual and quantitative predictors in the estimation of the disease progression and mortality of COVID-19 patients. They also showed that the time-to-event information calculated by pix2surv based on chest CT images enables stratification of the patients into low- and high-risk groups by a wider margin than those of the other predictors. Thus, pix2surv offers a promising approach for performing image-based prognostic prediction for the management of COVID-19 patients.