Matúš Falis

PhD Candidate at @BioMedAI_CDT.

Focus on analysis of biomedical text with the aid of ontologies.

T1D Cyborg

[Healtex @UK_healtex]

Updates on #HEALTAC2025: keynotes by Jason Fries (@jasonafries) and Alison O'Neil (@alisonQoneil) https://t.co/KjYbi6TUp0 #datasaveslives #NLProc

[Healtex @UK_healtex]

Updates on #HEALTAC2025: submission deadline extended to April 4th: https://t.co/KjYbi6TUp0 #datasaveslives #NLProc

[Healtex @UK_healtex]

Call for contributions for 8th annual UK healthcare text analytics conference (Glasgow, June 16-18, 2025): https://t.co/rCmPccIAlG #datasaveslives #NLProc #HEALTAC2025

[Healtex @UK_healtex]

Save the date: #HEALTAC2025 in Glasgow, June 16-18, 2025. We are delighted to announce the eighth annual UK healthcare text analytics conference. The call for contributions will be issued shortly. #datasaveslives #NLProc

[Aryo Pradipta Gema @aryopg]

LLMs are powerful, but we all know that they may generate hallucinations 🍄 even when answering trivial questions and solving simple tasks. To solve this, we propose DeCoRe – Decoding by Contrasting Retrieval Heads 🚀
🔗 Summary: https://t.co/go2oGWvmmC 🧵1/N

[Open Life Science AI @OpenlifesciAI]

🚨 Medical AI Research Alert! 🚨

Can GPT-3.5 transform the way we handle medical document coding? 🏥

@InfAtEd presents:
GPT-3.5 for Medical Discharge Summaries & ICD-10 Coding
Authors: @matt_falis ,@aryopg,@alexandrabirch1..

Here’s why it’s exciting: 1/ 👇🧵
#AI #MedicalAI #NLP #ICDCoding #MachineLearning #HealthTech #LLM #Bioinformatics #GenAI

[Matúš Falis @matt_falis]

@RamanDutt4 Congrats!

[Matúš Falis @matt_falis]

Many thanks to all the amazing experts in natural language processing and medicine who contributed to this study:
@aryopg, @ClementDong, @ljdaines, Siddharth Basetti, Michael Holder, @rosespenfold, @alexandrabirch1, @bea_alex

@BioMedAI_CDT, @acrcedincare
🧵 8/8

[Matúš Falis @matt_falis]

Can GPT-3.5 generate plausible clinical notes?
Generated discharge summaries have the correct diseases and procedures 🙂! but don’t tell plausible stories 🙁 . Though not authentic, the generated documents can augment ICD coding training!
👉 https://t.co/wqNEjhgmP6
🧵 1/8

[Matúš Falis @matt_falis]

GPT-3.5 can also code the generated text based on the standard descriptions in the prompt. For (securely communicated) real data the model needs to identify relevant concepts and provide the codes - performance drops dramatically. It's not viable for automated ICD coding.

🧵 7/8

[🧬Ian Simpson🏴󠁧󠁢󠁳󠁣󠁴󠁿 @tisimpson]

Currently recruiting for a PhD student in a collaboration with @Roche and Honghan Wu @UofGlasgow - https://t.co/EgwRk9CNLl as part of the @AI4BI_CDT using LLMs in healthcare.

[Aryo Pradipta Gema @aryopg]

I'm excited to announce that I’m starting my internship at @AstraZeneca today. I’ll be working on mitigating LLM hallucination in the clinical domain! 🍄
Reach out if you’re working on the same topic!! 🙌

[Matúš Falis @matt_falis]

@bea_alex @aryopg @ClementDong @ljdaines Okay, everything is in order. I shall sleep well tonight 😂

[Matúš Falis @matt_falis]

@bea_alex @aryopg @ClementDong @ljdaines You know, beyond that one picture you took of me explaining, I do not seem to be waving my hands around - at least there is less documentation of it. I wonder what went wrong with me over the past year.

[🧬Ian Simpson🏴󠁧󠁢󠁳󠁣󠁴󠁿 @tisimpson]

Applications are now open to join our new UKRI AI Centre for Doctoral Training in Biomedical Innovation for September 2024 entry. All the information you need to apply is ➡️ https://t.co/W7g5miBZ0j @UKRI_News @AI4BI_CDT