A radiologist in a large hospital sees hundreds of chest X-rays (CXRs) daily and must make quick and accurate diagnoses. The sheer volume of images is overwhelming, and there are not enough radiologists to analyze all the images promptly. This is where RadVLM comes into play.
RadVLM is an innovative computer program specifically designed to assist radiologists in interpreting chest X-rays. It can perform multiple tasks simultaneously: generating reports, detecting abnormalities, and interacting with doctors. The program was trained on over a million image-instruction pairs, enabling it to interpret and describe X-rays accurately.
RadVLM can answer questions and provide explanations, similar to a collegial exchange. For example, if a doctor asks, “What does this X-ray show?” RadVLM can provide a detailed answer and point to specific areas of the image.
The researchers hope that RadVLM will relieve radiologists and improve the quality of diagnoses. It could be particularly beneficial in hospitals with few radiologists or in regions with limited access to medical care.
For more information and detailed results, refer to the full article on arXiv: RadVLM: A Multitask Conversational Vision-Language Model for Radiology (published February, 5 2025).
Some of the authors are involved in the project “TRUST-RAD: Reliable AI assistance tools for radiology”, which is being funded in the 3rd Rapid Action Call.