AI can identify brain tumors almost as well as humans

 

In a new study, an AI model can distinguish healthy from cancerous brains with high accuracy.

The search for a brain tumor involves hours of analyzing images. But soon, radiologists may be able to count on technology to help them do this. Scientists have made progress in using artificial intelligence to read MRI scans. In tests, the model was able to detect cancer almost as well as humans.

AI radiologist can identify tumors with high accuracy

• In the same way that AI can already distinguish camouflaged animals in nature , scientists realized that the technology could differentiate cancerous cells that blend in with healthy ones.

• So they started training a model with publicly available MRI data from healthy and cancerous brains.

• Additionally, they used a knowledge transfer mechanism — when learning from one task can be reused in another — to improve the AI’s abilities.

• The tool's neural network proved to be almost as effective as human experts, delivering one or two false negatives in the diagnosis.

• The first network developed achieved an average accuracy of 85.99% in detecting brain cancer, while the other had an accuracy rate of 83.85%

  
  Images of different tumor types generated by multiple trained AI neural networks 

Transparency to support doctors

One of the most important features of the new AI radiologist is that it can explain its decisions clearly. After analysis, the model generates images highlighting specific areas to help doctors confirm their diagnoses. It’s almost as if the human expert has a third-party view of the scan results.

In the future, the idea is to make the technology’s transparency about its choices even more intuitive. This way, AI can be used as a reliable support in hospitals, for both doctors and patients.

Model still needs to be improved

Although it achieved satisfactory accuracy, the purpose of the AI ​​radiologist still needs improvement. Compared to human doctors, the model was 6% less accurate in detecting tumors. Much still needs to be studied before the network can be used in hospitals.

The main concern for scientists now is to make the technology’s processes even clearer. “AI can help with more accurate detection and diagnosis, but it is essential to understand how it does these analyses,” study author Arash Yazdanbakhsh explained to Medical Xpress .