One spine MRI scan, 76 diseases, and a draft report ready to check

Many AI models in medical imaging perform well on lab data but fail in real clinics. The focus has shifted from technical feasibility to clinical utility.

The radiology department at Peking University Third Hospital and United Imaging Intelligence developed a spine MR agent based on a one scan, many examinations principle. A single spine MRI produces a structured analysis of multiple conditions simultaneously. Prof. Yuan Huishu, head of diagnostics at the department, notes that the agent combines large model capabilities with actual clinical habits. The information is complete, accurate, and convenient. It removes repetitive observation, sorting, and text entry from the doctor's workload.

Scale of the burden

Spine MRI is one of the most common examinations in the imaging suite. At this Beijing hospital, it is the most frequent MR examination in the entire facility, averaging close to 300 scans per day. A report requires reviewing vertebrae, discs, ligaments, spinal cord, and nerve roots layer by layer, then describing herniations, canal stenoses, or changes in vertebral bodies. Without AI support, a report takes five to ten minutes on average from first draft to approval.

Prof. Yuan points out that traditional AI models solve one task or a few, acting like specialist observers of a single lesion type. They are often separate programs outside the PACS system. Doctors must transfer results and organize text manually, adding work instead of reducing it. Some tools generated answers inconsistent with the image, undermining trust in the technology.

Three thresholds

The agent makes three jumps. First, from recognizing a single disease to systematic assessment of many. It analyzes the image globally and outputs a report covering 76 diseases and abnormalities in one pass, evaluating spine sections sequentially. In a lumbar examination for back pain, it detects not only disc degeneration but also vertebral hemangiomas or compression fractures. Second, from marking foci to a full structured report: lesion detection, anatomical localization, symptom description, and the final structured report are generated together. Report writing efficiency increased by 56 percent. Third, from an external tool to an agent embedded in the AI PACS platform used daily by doctors.

The project formally started in July 2025. It followed an 18-month overhaul of the AI PACS platform by both parties. The team lists four technical challenges: precise anatomical localization of numerous spine sections, understanding radiologists' report-writing logic, the long tail of rare diseases with insufficient data, and avoiding model hallucinations. The agent already operates at the main hospital and four branches. AI remains an auxiliary tool; a radiologist must approve every result. The next stage is an external multicenter study.