Community Discussion · Policy

Testing Yinghe Yimai's Medical Imaging AI: Real Lab Insights After Two Days

Feng sirAug 102026/08/10 218 views

Let me clarify first, I'm not a doctor, I work in computer vision. Last week, while running model inference in the lab, I casually clicked on the demo entry for Yinghe Yimai's Medical Imaging AI 3.0. I intended to just take a quick look, but ended up spending the whole afternoon there. Later, I specifically scheduled a call with their product team. It took me two days in total to go through everything.

Policy

3 replies

Ctrl + Enter to reply

Jin XiujieAug 11

I didn't ask that much detail about the data sources, but they ran raw DICOM files, which should include device tags. I suggest you ask their product team directly. The image reconstruction algorithms for Philips and GE differ significantly. We've taken quite a few hits regarding model generalization in automotive projects.

SleepyAug 10

Here for anyone who has waited so long for segmentation models to load that they questioned their life choices... Your result in just over ten seconds already makes me jealous. As for the claim covering 94 diseases, waiting for follow-up real-world tests on edge cases. I'm genuinely curious how it distinguishes between micro-hemorrhages and calcifications.

Pan XuetingAug 10

Haha, I'm really skeptical about this "94 diseases" claim... I've tested other vendors before, their reports sounded amazing, but they fell apart immediately when given a different batch of data. The micro-hemorrhages and calcifications you picked are pretty tricky; these two are indeed easy to mess up on CTs. But if they passed your test, it means they do have some substance? BTW, did you test with data from their own devices or from another hospital? Images from different machines vary quite a bit—is this model still stable if switched to Philips or GE scanners?