PARIS – A consortium that includes Therapanacea SAS, GE Healthcare France SAS, Orange Healthcare SA, the AP-HP group of 39 hospitals in the Paris region and the French academic society for thoracic imaging has created a database for COVID-19.
Data collected from 10,000 computed tomography (CT) chest scans performed on patients suspected of having COVID-19 will help create a new generation of prognostic tools for this disease based on artificial intelligence (AI). “The objective is to develop predictive algorithms capable of quantifying the extent of lung lesions and relating these to other information on the patient’s state of health before infection,” Nikos Paragios, founder and CEO of Paris-based Therapanacea, told BioWorld.
Cross-referencing scan data with clinical annotations by radiologists
This French consortium is part of the STOIC project (thoracian scanner for the diagnosis of coronavirus-19 pneumonia), launched in March by Marie-Pierre Revel, head of the cardiothoracic imaging unit at Cochin Hospital. This large-scale project aims to analyze 10,000 chest scans performed for suspected COVID-19. “It's not just about collating scans of COVID-19 patients,” Revel said. The originality of the approach lies in the annotation of the scans by experts specializing in thoracic imaging, who will identify the anomalies linked to COVID-19, “and in the use of other information in the image on the patients’ state of health with regard to their coronary arteries, lungs, etc.” said Revel.
The Paris-area hospitals, plus those in Rennes, Lyon and Strasbourg, supplied the data. A group of 20 radiologists from the French Thoracic Imaging Society annotated the images. STOIC's approach relates to disease quantification and is based on automated CT, which uses deep learning and robust data-driven identification of physiologically inspired COVID-19 holistic patient profiling. “This will give radiologists a much more precise prognostic score for adapting medical care better,” said Paragios.
What’s more, identifying individuals who are most likely to deteriorate and require intubation “could accord efficient management of patient population and hospital resources, through anticipation regarding transfer to other hospitals and selection regarding patients for new drugs, thus reducing the need for intubation,” said Revel.
There are several moving parts within the consortium. Orange Healthcare SA is hosting the CT scan data in its secure cloud storage infrastructure. Meanwhile, the French division of GE Healthcare Inc., of Chicago, has deployed its Edison 3D image visualization platform, which allows remote annotation of lung lesions by segmenting zones affected by the virus. Thanks to Edison, which allows for the extraction of imaging markers for diagnosis and patient monitoring, “radiologists can also provide information on factors such as vascular, pulmonary or overweight comorbidity that may influence the course of COVID-19,” said Baptiste Perrin, director – imaging software R&D at GE Healthcare.
Therapanacea, a spinoff from the Centralesupélec engineering school, specializes in biomarkers and accompanying algorithms for prognosis in oncology. It has developed a model integrating AI to quantify COVID-19 automatically and establish a short-term prognosis for patients based on their lung scans upon admission to the hospital.
A retrospective, multicenter study has been carried out on data from scans originating from a cohort of 693 patients with COVID-19, or 21,360 CT slices, and from eight large university hospitals in France. Preliminary results reveal that the AI solution assesses the severity of lung damage within minutes with the same accuracy as an experienced radiologist. The software suite has a success rate of 75% for predicting the severity of lung damage, vs. 60% for an experienced radiologist. This represents a 10% gain in the relevance of the prognosis for COVID-19 pneumonia and in the implementation of more effective therapeutic strategies.
“We are currently continuing the study in order to validate these results using a larger cohort of hospitals,” said Paragios, who added that this prognostic software suite is expected to obtain CE marking approval based on its use for research. GE Healthcare plans to distribute this application within its portfolio devoted to the diagnosis of COVID-19.