A Reality Check of Vision-Language Pre-training in Radiology: Have We Progressed Using Text?

Keyphrases

• Vision-language Models 100%
• Reality Check 100%
• Text Supervision 66%
• Large-scale Data 33%
• Medical Image Analysis 33%
• Expert Knowledge 33%
• Radiology 33%
• Medical Dataset 33%
• Text-to-image 33%
• Heterogeneous Data Sources 33%
• Step-back 33%
• Open-vocabulary 33%
• Vision-language Pre-training 33%
• Medical Concepts 33%
• Noisy Text 33%
• Rich Feature Representation 33%

Computer Science

• Language Modeling 100%
• Data Source 66%
• Recent Literature 33%
• Large Scale Data 33%
• Image Analysis 33%
• Expert Knowledge 33%
• Heterogeneous Data 33%

Social Sciences

• Language Modeling 100%
• Expert Knowledge 33%