Deep Learning Demonstrates Potential for Lung Cancer Detection in Chest Radiography
Menée à partir de 10 285 clichés radiographiques du thorax réalisés sur 10 202 personnes (âge moyen : 54 ans ; 5 857 hommes) puis validée à partir de 100 525 autres clichés radiographiques réalisés sur 50 070 personnes supplémentaires (âge moyen : 53 ans ; 28 090 hommes), cette étude évalue, par rapport aux interprétations de radiologues, la performance d'un algorithme d'apprentissage automatique pour détecter un cancer du poumon
The decades-old concept of computer-aided diagnosis (1) is alive and well—but transformed by accelerating advances in machine learning and artificial intelligence (2). These advances have created a virtual explosion of academic, clinical, and corporate researchers who are contributing to the development and testing of radiologic decision support systems. The ultimate goal of these systems is clinical deployment after meeting regulatory requirements. Two almost-certain aspects of regulatory compliance include demonstrating a level of performance at least comparable with experienced radiologists (1) and validation on an extensive set of unbiased clinical data (2). In this issue of Radiology, Lee et al (3) have achieved success in both of these real-world scenarios.
Radiology , éditorial, 2019