New AI Can Grade CT Scans Like a Radiologist—But With Explanations
This could lead to safer, lower-dose CT scans and faster diagnoses for you.
Researchers propose a measurement-aware multimodal LLM, built on Qwen3.8-27B, that predicts the score radiologists would assign to a low-dose CT slice without a full-dose reference—as needed to compare dose protocols and denoisers. It rates a slice from its image plus six image-domain measurements of noise, sharpness, high-frequency content and streaks, given to the language model as percentiles in text.
On the LDCT-IQA 2023 test set, it reaches an Overall score of 2.735, on par with the retrained challenge winner (2.735) and MedSigLIP-FiLM (2.713), and substantially above its base MLLM used as a zero-shot agent. Shuffling the measurements at test time reduces the score by 0.048, mostly through noise, but the same measurements also lead the model to assign higher scores to over-smoothed denoised images.
- The AI grades low-dose CT scans as accurately as top methods but also explains why it gave a certain score.
- It uses simple measurements like noise and sharpness to make judgments, which helps it be transparent.
- A limitation: it sometimes prefers overly smoothed images, which could hide important details.
Why It Matters
Safer, lower-dose CT scans with reliable quality checks could mean less radiation for patients and faster diagnoses.