@frazierhuo.bsky.social
[4/4] 🙌 Huge thanks to my amazing co-leads @jason-fries.bsky.social ,
@Alejandro Lozano
& co-authors @Jeya_Valanarasu,
@Ethan Steinberg
, Louis Blankemeier, Akshay Chaudhari, Curtis Langlotz, Nigam Shah!
📂 Dataset:
🩺 EHR: redivis.com/datasets/dzc...
🩻 Imaging: aimi.stanford.edu/datasets/ins...
@Alejandro Lozano
& co-authors @Jeya_Valanarasu,
@Ethan Steinberg
, Louis Blankemeier, Akshay Chaudhari, Curtis Langlotz, Nigam Shah!
📂 Dataset:
🩺 EHR: redivis.com/datasets/dzc...
🩻 Imaging: aimi.stanford.edu/datasets/ins...
INSPECT EHR
The INSPECT dataset (Integrating Numerous Sources for Prognostic Evaluation of Clinical Timelines) contains de-identified longitudinal electronic health records (EHRs) from a large cohort of pulmonary...
redivis.com
February 2, 2025 at 6:17 AM
[4/4] 🙌 Huge thanks to my amazing co-leads @jason-fries.bsky.social ,
@Alejandro Lozano
& co-authors @Jeya_Valanarasu,
@Ethan Steinberg
, Louis Blankemeier, Akshay Chaudhari, Curtis Langlotz, Nigam Shah!
📂 Dataset:
🩺 EHR: redivis.com/datasets/dzc...
🩻 Imaging: aimi.stanford.edu/datasets/ins...
@Alejandro Lozano
& co-authors @Jeya_Valanarasu,
@Ethan Steinberg
, Louis Blankemeier, Akshay Chaudhari, Curtis Langlotz, Nigam Shah!
📂 Dataset:
🩺 EHR: redivis.com/datasets/dzc...
🩻 Imaging: aimi.stanford.edu/datasets/ins...
[3/4] 🛠️ 𝗛𝗼𝘄 𝗜𝘁 𝗪𝗼𝗿𝗸𝘀
💠 Transform EHR timelines into TTE tasks, predicting event time distributions.
💠 Train a 3D vision encoder to extract imaging biomarkers for survival analysis.
💠 Enables AI to model long-term health trajectories for better risk stratification.
💠 Transform EHR timelines into TTE tasks, predicting event time distributions.
💠 Train a 3D vision encoder to extract imaging biomarkers for survival analysis.
💠 Enables AI to model long-term health trajectories for better risk stratification.
February 2, 2025 at 6:13 AM
[3/4] 🛠️ 𝗛𝗼𝘄 𝗜𝘁 𝗪𝗼𝗿𝗸𝘀
💠 Transform EHR timelines into TTE tasks, predicting event time distributions.
💠 Train a 3D vision encoder to extract imaging biomarkers for survival analysis.
💠 Enables AI to model long-term health trajectories for better risk stratification.
💠 Transform EHR timelines into TTE tasks, predicting event time distributions.
💠 Train a 3D vision encoder to extract imaging biomarkers for survival analysis.
💠 Enables AI to model long-term health trajectories for better risk stratification.
[2/4] 🔥 Key Highlights
✅ 𝗧𝗧𝗘 𝗽𝗿𝗲𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴: Predicts time until critical events from 3D imaging.
✅ 𝗠𝗮𝘀𝘀𝗶𝘃𝗲 𝘀𝗰𝗮𝗹𝗲: 8,192 tasks, 18,945 CTs, 225M EHR events—largest EHR+3D imaging dataset.
✅ 𝗣𝗿𝗼𝗴𝗻𝗼𝘀𝘁𝗶𝗰 𝗯𝗼𝗼𝘀𝘁: +23.7% AUROC, +29.4% Harrell’s C, +54% better calibration w/o losing diagnostic accuracy.
✅ 𝗧𝗧𝗘 𝗽𝗿𝗲𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴: Predicts time until critical events from 3D imaging.
✅ 𝗠𝗮𝘀𝘀𝗶𝘃𝗲 𝘀𝗰𝗮𝗹𝗲: 8,192 tasks, 18,945 CTs, 225M EHR events—largest EHR+3D imaging dataset.
✅ 𝗣𝗿𝗼𝗴𝗻𝗼𝘀𝘁𝗶𝗰 𝗯𝗼𝗼𝘀𝘁: +23.7% AUROC, +29.4% Harrell’s C, +54% better calibration w/o losing diagnostic accuracy.
February 2, 2025 at 6:12 AM
[2/4] 🔥 Key Highlights
✅ 𝗧𝗧𝗘 𝗽𝗿𝗲𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴: Predicts time until critical events from 3D imaging.
✅ 𝗠𝗮𝘀𝘀𝗶𝘃𝗲 𝘀𝗰𝗮𝗹𝗲: 8,192 tasks, 18,945 CTs, 225M EHR events—largest EHR+3D imaging dataset.
✅ 𝗣𝗿𝗼𝗴𝗻𝗼𝘀𝘁𝗶𝗰 𝗯𝗼𝗼𝘀𝘁: +23.7% AUROC, +29.4% Harrell’s C, +54% better calibration w/o losing diagnostic accuracy.
✅ 𝗧𝗧𝗘 𝗽𝗿𝗲𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴: Predicts time until critical events from 3D imaging.
✅ 𝗠𝗮𝘀𝘀𝗶𝘃𝗲 𝘀𝗰𝗮𝗹𝗲: 8,192 tasks, 18,945 CTs, 225M EHR events—largest EHR+3D imaging dataset.
✅ 𝗣𝗿𝗼𝗴𝗻𝗼𝘀𝘁𝗶𝗰 𝗯𝗼𝗼𝘀𝘁: +23.7% AUROC, +29.4% Harrell’s C, +54% better calibration w/o losing diagnostic accuracy.