medRxiv 预印本· Fathi Kazerooni, A., Batal, F., Vunnava, B., Damaraju, E., Tavallaii, A., Gandhi, D., Gottipati, A., Zhu, P., Chrysochoou, D., Varshochi, S., Adib Moradi, S., Familiar, A. M., Chiang, A., Vaghani, Y., Azemi, G., Doig, M., To, M.-S., Luck, A., Kim, M. C., Khalili, N., Khalili, N., Jain, P., Lee, C., Hansford, J. R., Di Ieva, A., Storm, P. B., Resnick, A. C., Reddy, K., Vossough, A., Nabavizadeh, A. ·· 20 天前AI 评分33
PedDx:基于T2加权MRI的儿童后颅窝肿瘤诊断端到端深度学习框架及智能体AI独立外部评估
PedDx: An End-to-End Deep Learning Framework for Pediatric Posterior Fossa Tumor Diagnosis with Agentic AI-Enabled Independent External Evaluation
AI 导读
PedDx是一个无需分割的儿童后颅窝肿瘤诊断深度学习框架,基于T2加权MRI与预训练3DINO表征,并采用校准预测和基于置信度的弃权机制。在114例复制集和129例外部队列中,分类准确率分别为79.8%和81.4%,剔除弃权病例后升至90.5%和88.7%。外部机构独立执行完整智能体工作流,对40例独立测试病例的准确率达92.5%,非弃权病例为97.3%。
来源:medRxiv 预印本 · medrxiv.org