New AI Checks Heart Scan Quality to Make Treatments Safer

For people with atrial fibrillation (an irregular heartbeat), doctors often use a type of MRI called LGE-MRI to find scar tissue in the heart. That scar tissue is the target for a procedure called ablation, which uses heat or cold to fix the heartbeat. But the MRI needs to be clear. If it's blurry or has noise, the doctor might burn the wrong spot — which is dangerous.

Right now, a radiologist looks at the images and makes a judgment call about whether they're good enough. That's subjective, and it's hard to do at large scale. Different radiologists might make different calls, and there's no written record of why a scan was rejected. This study proposes an automated system using vision language models — AI that can 'see' images and describe them in text.

The system works in two steps. First, an AI looks at the MRI and writes a short radiology-style report, checking five things: noise, motion blur, boundary clarity, tube (pulmonary vein) visibility, and whether parts are cut off. Then, a second AI reasoning step turns that report into two things: a quality score and a simple yes/no about whether the scan can be used for planning.

The researchers tested it on 60 images from 20 patients. One AI called InternVL2 was best at scoring the individual quality criteria (65% accuracy), and another called DeepSeek got the 'usable or not' decision perfectly right every time. The sample is small, but the results show this could work. If proven on larger groups, this could give hospitals a consistent, automatic safety check before every ablation procedure — catching bad scans early and protecting patients.