Back to list
This article was auto-translated.View original (中文)
Tech1mo ago

New Version of DeepSeek Can Understand CT Scans

Only 5 days after the release of DeepSeekV4, new features are appearing almost daily. Researchers previewed multimodal capabilities yesterday, and a grayscale test is already available. Many users have discovered an image recognition mode on the DeepSeek web interface, meaning it can now understand image information. While this ability won't directly enhance AI's programming or reasoning performance, it will be very convenient to use. You can simply upload screenshots of problems you encounter, and DeepSeek can analyze them itself, which is easier than describing the problem yourself.

New Version of DeepSeek Can Understand CT Scans

Users with access to the grayscale test have even used professional images, such as CT scans taken in hospitals, to test DeepSeek's image recognition capabilities, and were amazed by the results.

The CT image uploaded by @砖头 from the Linux.do community is from a professional paper. After DeepSeek analyzed it, it could accurately identify the content of the image and perform professional analysis, ultimately providing several possible diagnoses, including several different types of pneumonia.

The paper containing this CT scan has a clear conclusion. Comparing the two, it can be seen that DeepSeek's analysis is quite reliable, and it can play the role of an AI doctor in this regard.

However, AI is still AI. It can help everyone analyze situations, but major medical examinations and disease confirmations still require analysis and confirmation by hospitals and doctors.

If it's not a serious illness, AI can now be used as a doctor for ordinary medical problems. There are now many AI applications trained by professional medical large models. They are sufficient to judge problems and provide suggestions for everyday issues, and there is no need to go to the hospital and queue for minor problems.

Returning to the issue of DeepSeek, they have also conducted research in the field of multimodality before, and their open-source OCR technology has even reached the world's top level. Therefore, their visual capabilities are also worth looking forward to, and can further enhance the capabilities and usage limits of the DeepSeek V4 large model.