Stanford AIDE

@StanfordRad group dedicated to improving the safety, reliability, and equity of AI in healthcare, co-directed by @larson_david_b and @Dr_ASChaudhari

[Stanford AIDE @stanford_aide]

Unlike traditional task-specific AI, foundation models learn from massive, multimodal, unlabeled datasets, opening big possibilities in radiology (and challenges too).
Magda Paschali, PhD & Akshay Chaudhari, PhD share key insights on the @RSNA podcast

[Stanford AIDE @stanford_aide]

Hospitals vary widely in their patient populations, imaging protocols, and workflows. An AI model trained in one environment may not generalize well to another. Without rigorous evaluation, it is uncertain whether AI tools will perform as expected in new deployment settings.

[Stanford AIDE @stanford_aide]

This summer, we’ve welcomed a few student researchers to Stanford’s AIDE Lab

They’re tackling real-world challenges in AI & healthcare from radiograph evaluation models to boosting AI performance without retraining.

Meet Samhita, Alara & Nathan!

[Stanford AIDE @stanford_aide]

𝗜𝗳 𝘆𝗼𝘂 𝗮𝗿𝗲 𝗶𝗻𝘁𝗲𝗿𝗲𝘀𝘁𝗲𝗱 𝗶𝗻 𝗰𝗼𝗻𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗻𝗴 𝘁𝗼 𝘁𝗵𝗶𝘀 𝗶𝗺𝗽𝗼𝗿𝘁𝗮𝗻𝘁 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵, 𝗽𝗹𝗲𝗮𝘀𝗲 𝗲𝗺𝗮𝗶𝗹 [email protected] and we will provide additional details! 😃

[Stanford AIDE @stanford_aide]

We are excited to invite you to participate in an online research study at @StanfordMed focused on improving patient-centered communication in healthcare 🩻 This study involves using a grading scheme designed to assess the quality of patient-friendly radiology reports 🗒️

[Stanford AIDE @stanford_aide]

Your perspective will provide valuable insights that help ensure the grading scheme is effective and easy to use. Participation is entirely voluntary, and all responses will be kept confidential.

[Enzo Ferrante @enzoferrante]

Falta una semana para @Khipu_AI , la mayor conferencia de AI de LATAM, que este año tendrá lugar en Santiago de Chile!

Incluso si no quedaste seleccionado/a, hay varios eventos abiertos al público Jueves y Viernes a los que podés sumarte con registro previo.

Toda la info aca👇

[Stanford AIDE @stanford_aide]

Joint work led by @larson_david_b, @koiralaaaa, Lina Y. Cheuy, @magdapasc, Dave Van Veen, Hye Sun Na, Matthew B. Petterson, @zhongnanf and @Dr_ASChaudhari

[Stanford AIDE @stanford_aide]

Also read our features in
👉 @AuntMinnieEuro: https://t.co/AndKA70zOU
👉 The @RSNA 2024 Daily Bulletin: https://t.co/kftv7s3jTB

[Stanford AIDE @stanford_aide]

Read the full paper 👉 https://t.co/3RiqrYkR4i
Evaluate your own clinical histories 👉 https://t.co/N7990dgMRK
Check out our open-source code and model 👉 https://t.co/jlsUTHtl8f

[Stanford AIDE @stanford_aide]

Our method provides a way to reliably compare and assess the quality of clinical histories at scale and highlights the overall potential of AI tools for quality assessments of text

[Stanford AIDE @stanford_aide]

We show that LLMs can assess the quality of clinical histories at scale:
💡 Our fine-tuned Mistral-7B model achieved 91% accuracy, matching or even surpassing inter-radiologist agreement
💡Applied to 48,942 clinical histories, only 26.2% contained all five essential elements

[Stanford AIDE @stanford_aide]

Radiologists depend on concise, relevant, and complete clinical histories to accurately interpret imaging studies and provide a more useful radiology report. However, the clinical histories accompanying imaging orders are often incomplete, inconsistent, or missing key details

[Stanford AIDE @stanford_aide]

Can we use open-source LLMs to assess the quality of clinical histories? 🤔 Our latest results, published in @radiology_rsna show that we can!

[Radiology @radiology_rsna]

Producing a benchmark for completeness in extracted clinical history, an easily deployable fine-tuned open-source LLM rivaled GPT-4 Turbo in performance, with substantial agreement with radiologists. @magdapasc @Dr_ASChaudhari https://t.co/KD1u76FvpX

[Radiology @radiology_rsna]

Producing a benchmark for completeness in extracted clinical history, an easily deployable fine-tuned open-source LLM rivaled GPT-4 Turbo in performance, with substantial agreement with radiologists. @magdapasc @Dr_ASChaudhari https://t.co/KD1u76FvpX