Two variables are sufficient to provide reliable risk stratification of hepatic decompensation and death for cirrhosis patients. See our recent publication.

Thanks to the authors for making the 100th citation of RadBERT. Multi-domain improves classification in out-of-distribution and data-limited scenarios for... - #ScholarAlerts

It's my honor that my special topic course #DeepLearning for #Neuroscience was voted by graduate students as their "preferred topic" and I was voted as a "preferred instructor" unanimously. Really appreciate that. 🤓

@Radiology_AI A.5 The papers that I mentioned earlier are a good references :) #RadAIChat

@Radiology_AI A.4 It’s difficult to apply commonly used explanation generation tools such as SHAP to a BERT-based model, though solutions have been proposed to address this issue. Then again this is critical when applying an AI model into clinical practice. #RadAIChat

@Radiology_AI With this capability, we can use BERT to ensure that LLM doesn’t generate undesirable sentences. See also presented in RSNA 2023. #RadAIChat

@Radiology_AI I have a paper back in 2008 shows that reading English backward always leads to better result than reading forward for the task of tagging gene mentions in a sentence and combining bidirectional reading leads to the best result. #RadAIChat