An expert-level generalist AI for abdominal CT diagnosis
阿里巴巴達摩院與浙江大學醫學院附屬第一醫院等研發出通用醫療影像AI模型DAMO RADAR，登上國際頂級學術期刊《科學》(Science)。
模型面向腹部增強CT(電腦掃瞄)診斷，可一次性識別超過146種病症，其準確性首次達到影像科專家級水平，現已正式開源。
據悉，達摩院長期投入醫療AI，用AI識別CT影像中肉眼難以識別的微小病灶，先後突破了胰腺癌、胃癌、腸癌等消化系統腫瘤篩查，以及主動脈夾層預警，三年內斬獲5篇《自然·醫學》和10多篇其餘頂刊論文。
達摩院此次推出並開源了多場景、多病種的通用模型DAMO RADAR，樹立了醫療影像AI的全新里程碑，有望推動醫療行業向通用人工智能(AGI)時代邁進。
airbus330 wrote:質疑《科學》(Science)?
一次性 ？(恕刪)
還好沒有說抄襲，說蒸餾、說偷問美國AI要答案
RESULTS
We constructed RAD-CT, a large abdominal CT dataset, consisting of 424,911 examinations and 15 million anatomy-wise pairs for fine-grained, large-scale training. Evaluated across 18 anatomical structures and 146 imaging findings, RADAR achieved an area under the receiver operating characteristic curve (AUC) of 0.913 (95% confidence interval, 0.911 to 0.915) for the diagnosis of 146 findings in a real-world internal cohort of 39,160 examinations and AUCs of 0.874 to 0.912 across eight external centers. In pathology-confirmed evaluation for four types of cancers (liver, pancreas, stomach, and colorectum), performance reached AUCs of 0.891 to 0.984. In another cross-population cohort, RADAR achieved an AUC of 0.883 without any fine-tuning. In acute abdominal conditions, which had been excluded from initial training, RADAR maintained high performance with an AUC of 0.904, underscoring strong generalization to urgent, unfamiliar scenarios. In a reader study with 26 radiologists from 14 centers, RADAR outperformed most participants and increased sensitivity by ~10% in collaborative use. RADAR also provided attention maps to highlight diagnostic-relevant visual cues, thereby facilitating better interpretation, a feature of considerable value in clinical practice.
內文搜尋