Researchers at Stony Brook University are working on an AI-based method that could change the landscape of medical imaging by making CT scans faster, safer, and more accurate. This new technique does not require extensive training data or expose patients to high doses of radiation.
The research is supported by Stony Brook University’s AI Seed Grant Program. The team is utilizing a deep learning technique known as Deep Image Prior (DIP). Unlike traditional AI tools, DIP constructs high-quality images from scratch using only the scan data available at the moment. This allows it to function effectively even with limited data or when a patient's medical history is unavailable.
Ziyu Shu, a senior postdoctoral associate in the Department of Radiation Oncology, has developed a new method called RBP-DIP (Residual Back Projection with Deep Image Prior). This approach employs a step-by-step process to enhance image clarity, particularly in challenging situations such as patient movement during scans or when minimizing radiation exposure is necessary.
The team is led by Principal Investigator Xin Qian, clinical assistant professor in the Department of Radiation Oncology. They are addressing challenges in current medical imaging systems and have already observed promising results. By using experimental CT machines, they have successfully created clear images with significantly fewer X-ray angles than usual, avoiding grainy artifacts or blurring common with older methods. In some instances, just 51 projections were sufficient to produce a sharp and detailed scan.
For further details, read Aknita Nagpal's full story on the AI Innovation Institute website.