Happy to introduce AMIE (Articulate Medical Intelligence Explorer) our research LLM for diagnostic conversations. AMIE surpassed Primary Care Drs in conversational quality & diagnostic accuracy in a "virtual OSCE"-style randomized study. Preprint ➡️ arxiv.org/abs/2401.05654 (1/7)

Ten months ago, we launched the Vesuvius Challenge to solve the ancient problem of the Herculaneum Papyri, a library of scrolls that were flash-fried by the eruption of Mount Vesuvius in 79 AD. Today we are overjoyed to announce that our crazy project has succeeded. After 2000

Big dataset of neck CTs+radiotherapy structure segmentations just released by Scott Bratman and colleagues through TCIA, >3000 patients. Great to see the commitment to open science, people will do cool things with this. aapm.onlinelibrary.wiley.com/doi/full/10.10…

Love a simple+elegant ML method. Vision Transformer may be the archetype. Divide image into rectangular patches, flatten each, pass them in sequence into a vanilla transformer designed for language, train on a ton of data, achieve SOTA for classification. Other examples?

Excited about vision-language models for radiology. Ground floor: leverage LLMs’ world knowledge to interpret+manipulate scans. Advanced: Give the models visual understanding so they can learn spatial relationships (imagine a LLM reading radiology textbooks to learn anatomy).

I never quite understood the "heterogeneity of treatment effect is rare/hard to prove" side. In oncology there are many examples of drugs that work better for certain cancer sub-types. For one dramatic example, this from Mok et al. NEJM 2009:

Exciting to see new trial results in JAMA Onc from Manali Patel MD & co., studying a lay health worker-led advance care planning intervention in a Stanford oncology clinic. Increased provider ACP documentation, palliative care and hospice use. jamanetwork.com/journals/jamao…

Sharing our new preprint on external beam radiation for neuroendocrine tumors. Led by the brilliant Julie Koenig, MD. Had trouble finding a journal but we felt important to get the data out there because disproves the common belief that RT not effective. medrxiv.org/content/10.110…

Cool JCO study from Kerry Reynolds and colleagues showing that LLM can accurately identify immune related adverse events from immunotherapy. And, impressive manual review of 8,800 hospital admission records (!!). ascopubs.org/doi/full/10.12…