CSIRO Researchers Trial AI System to Automate Chest X-Ray Reporting

Artificial intelligence is transforming medical image analysis, with researchers at CSIRO’s Australian e-Health Research Centre (AEHRC) leading the charge. Dr. Aaron Nicolson, a research scientist at AEHRC, is developing a visual language model (VLM) to automate the reporting of chest X-rays—a task traditionally performed by radiologists. Chest X-rays are vital for diagnosing conditions like pneumonia, heart disease, and lung cancer, as well as checking the placement of medical devices. However, Australia faces a growing shortage of radiologists, making it difficult to keep up with rising patient demand, especially as the population ages. Nicolson’s goal is to create AI that supports radiologists by generating accurate, timely reports without replacing them. The VLM works by analyzing chest X-ray images alongside patient information such as referral notes. It learns to produce radiology reports by studying thousands of paired images and expert-written reports. The more data the model processes, the better it becomes at generating accurate descriptions and findings. Nicolson took the project a step further by including additional clinical data—such as emergency department records, vital signs, medications, and patient complaints—into the model’s input. This extra context significantly improved the accuracy of the generated reports, bringing the technology closer to being ready for real-world use. The team is now conducting a trial with Princess Alexandra Hospital in Brisbane, comparing AI-generated reports with those from human radiologists. They are also seeking more clinical partners to expand testing across diverse healthcare settings. Meanwhile, Dr. Arvin Zhuang, a postdoctoral researcher at AEHRC, is exploring other uses of VLMs, such as extracting information from medical documents treated as images rather than text. This approach can speed up data retrieval and improve efficiency in clinical workflows. Ethical considerations remain central to the research. Nicolson stresses the importance of ensuring the AI is fair and effective across all patient groups by addressing potential biases in training data. He emphasizes that the technology is not meant to replace doctors but to assist them, with radiologists always involved in final decision-making. As AI continues to evolve, projects like these show how advanced models can enhance healthcare delivery—making diagnostics faster, more accessible, and more reliable—while maintaining the essential human oversight required in medicine.