Endoscopic screening for oesophageal cancer: empowering artificial intelligence with a high-quality examination
Mené en Chine sur 11 846 personnes, cet essai multicentrique évalue la performance d'un système d'endoscopie assistée par l'intelligence artificielle pour détecter des lésions pré-cancéreuses et des carcinomes épidermoïdes superficiels de l'oesophage
Oesophageal squamous cell carcinoma, the predominant oesophageal cancer subtype in the world, has a poor prognosis when diagnosed at a late stage. 1 Data to support the practice of endoscopic screening for oesophageal squamous cell carcinoma in populations at risk come from prospective cohort studies and randomised controlled trials that show a significant improvement in early detection and treatment, with an associated reduction in all-cause mortality. 2 A high-quality endoscopic screening examination for oesophageal squamous cell carcinoma involves systematic visualisation of the entire oesophagus, preferentially with the use of high-definition white light endoscopy in combination with digital chromoendoscopy. Unfortunately, in clinical practice, endoscopic screening for oesophageal squamous cell carcinoma is associated with a misdiagnosis rate of 6·4% (95% CI 4·4–9·3), in part due to the subtle appearance of early lesions. 3 Strategies to improve diagnostic performance include the use of iodine staining, a technique associated with longer procedural times and a slightly higher risk of adverse events compared with standard-of-care endoscopy. 4 The use of magnified endoscopy in combination with digital chromoendoscopy has also been shown to improve the diagnosis of early oesophageal squamous cell carcinoma but requires special equipment and technical expertise. 4 Advances in artificial intelligence (AI) have led to the development of advanced computer-aided detection algorithms, which promise to improve early detection of oesophageal squamous cell carcinoma and reduce lesion misdiagnosis. An advantage of computer-aided detection systems compared with adjunct screening modalities is their ability to provide automated real-time lesion identification with minimal training for use. Several published computer-aided detection models show high accuracy in detecting early-stage oesophageal squamous cell carcinoma, but few have truly been put through rigorous testing.
The Lancet Gastroenterology & Hepatology , commentaire, 2022