Outpatient imaging report turnaround times can matter... Shown here: splenic infarct and LV thrombus on a single image.

Default GPT-3.5 and GPT-4 were accurate at answering radiology board multiple-choice questions but had poor repeatability and robustness and were frequently overconfident, limiting usability without domain-specific optimization. bit.ly/3wJrEdZ

Large Language Models for Automated Synoptic Reports and Resectability Categorization in Pancreatic Cancer | Radiology pubs.rsna.org/doi/10.1148/ra…

🧵 Structured radiology reports for pancreatic ductal adenocarcinoma (PDAC) helps surgical planning, yet it is not universally used by Rads. Can large language models (LLMs) help? A new study from Radiology explored this further #RadInTraining #TWEETORIAL Rad In Training Editor

Large #Language #Models for Automated #Synoptic Reports and #Resectability Categorization in #PancreaticCancer | Radiology pubs.rsna.org/doi/10.1148/ra…

GPT-4 enabled near-perfect creation of pancreatic ductal adenocarcinoma synoptic reports from original reports and accurate resectability categorization using chain-of-thought prompting. University Medical Imaging Toronto (UMIT) Rajesh Bhayana Satheesh Krishna bit.ly/4cqQ8I2

3/ On the other hand, such criticism of the limited evaluations of LLMs in the radiology AI literature is valid. This Radiology paper by Rajesh Bhayana et al. shows the limitations of simple prompts and how one can improve LLM performance in complex medical tasks.

Doctors please remember For you it's just another appointment. Just another rounds visit. But for patients those 10 minutes are the most important visit of their day. Sometimes of their lives.

AI using retrieval-augmented generation, which prioritizes authoritative sources, outperforms conventional large language models on radiology board–style exams. UofT Medical Imaging @ImagingToronto Rajesh Bhayana Aly Fawzy Satheesh Krishna bit.ly/3TWOAPt

🚀New Article: Retrieval Augmented Generation (RAG) for LLMs is elevating radiology board-exam performance! Discover how generative AI is shaping the future of medical education. #AI Rajesh Bhayana Satheesh Krishna @ImagingToronto Aly Fawzy UofT Medical Imaging pubs.rsna.org/doi/10.1148/ra…

Congratulations to our Dr. Rajesh Bhayana His Chatbots and Large Language Models in Radiology: A Practical Primer for Clinical and Research Applications was named a top 10 publication in Radiology in 2024: ✅2nd in citations ✅ 5th in downloads! ➡️tinyurl.com/2nhks5uw

Google Gemini 2.0 Flash Realtime performs exceptionally well on this CT Abdomen case -- it answered my questions right and nailed the diagnosis (acute pancreatitis), including some complications to look for. Very impressive!! Imagine a version of this tuned specifically for

PSA: General-purpose multimodal models, like Gemini 2.0, Grok-2, and GPT-4o are NOT good at interpreting medical images. This is expected, as they were not trained for this purpose. Last week, I shared a video of Gemini 2.0 Realtime looking at a CT that went viral. It was an

Coming in at no. 2️⃣ most cited Radiology article published this year: check out this user-friendly introduction to the research and clinical applications of LLMs and AI in radiology by Dr. Rajesh Bhayana, with helpful diagrams! bit.ly/3ZCsRyl

Large language models enabled the creation of automated clinical histories for oncologic imaging, offering more complete and radiologist-preferred requisition histories for improved image interpretation and safety. Rajesh Bhayana #Radiology #AI pubs.rsna.org/doi/10.1148/ra…

LLMs help solve an age-old problem in Radiology: lack of clinical history! In our study, AI generated histories were far more complete than those provided by physicians — capturing primary cancers, acute symptoms, and relevant surgeries much more often. Radiologists

Straight-out-of-the-box #LLMs still struggle with radiology text classification. Check out how Rajesh Bhayana and colleagues used a hybrid LLM approach for O-RADS score calculation from MRI reports! shorturl.at/J2Xq6 #RadInTraining Victoria Chernyak, MD MS FSAR FESGAR FACR Rad In Training Editor Radiology

Can AI outscore radiologists on O-RADS MRI? A hybrid LLM-based app by Rajesh Bhayana et al combined feature classification with deterministic elements to assign scores from adnexal lesion descriptions, and it outperformed original reports! #Radiology pubs.rsna.org/doi/10.1148/ra…

"AI isn't replacing radiologists" good article Expectation: rapid progress in image recognition AI will delete radiology jobs (e.g. as famously predicted by Geoff Hinton now almost a decade ago). Reality: radiology is doing great and is growing. There are a lot of imo naive