AIMA 2026 projects ✦ Project V ✦ Surgical AI
AI that understands the cardiac operating room
Cardiac surgery is a team doing high-stakes steps under time pressure, while the patient’s circulation shifts minute by minute. We want AI that understands all of it: how the patient will respond, how the team communicates, and how each step is performed, and then links those moments to what happens to the patient afterwards.
In partnership with
One case, three tracks, one clock
Schematic: the waveform, speech turns and video events are illustrative, not patient data.
Why the operating room
- Skill seen on video predicts outcomes: patients of surgeons in the bottom skill quartile had 14.5% complications, against 5.2% in the top quartile1.
- About 30% of team exchanges failed in one observational study, and a third of those had visible effects2. An OR black box recorded a median of 20 errors per laparoscopic case3.
- Waveform AI predicts hypotension 5 minutes ahead with AUC 0.974, and in a trial it cut hypotension time from 32.7 to 8.0 minutes5.
The gap we target
These tools were built for noncardiac surgery, not the switch from the heart-lung machine back to the heart, where post-bypass vasoplegia still lacks a uniform definition6. Team skills are still rated by human observers7. And we know of no public cardiac-surgery video dataset. The project’s planned cohort of 300 operations at University Medical Center Ho Chi Minh City and Hue Central Hospital, linked to outcomes up to one year, is built to fill that gap.
Physiological digital twin
Track each patient’s physiological state and forecast hemodynamics through weaning from cardiopulmonary bypass, with early warning of instability. Prototype on public intraoperative signals8.
OR black box
Combine audio and video to see who says what, who does what, and when: speaker diarization, person tracking and action recognition, then coordination problems linked to delays and complications.
CardioSurgVal
Recognize anatomy, instruments and steps in cardiac surgical video, retrieve clips as evidence with vision-language models, and predict complications and recovery up to one year.
Data
Skills by track
- Twin: PyTorch, signal processing, time-series forecasting, uncertainty; hemodynamics a plus
- Black box: speech, diarization, tracking, action recognition
- Video: surgical video, vision-language models, clinical annotation
Good fit if you
- Can commit six months
- Want to work where engineering meets the operating room
- Are a clinician or student who can annotate cardiac surgery
Physiology
Signals, forecasting and early warning.
Speech & team
Transcripts, diarization, call-out detection.
Surgical vision
Steps, instruments and evidence clips.
Outcomes
Linking all tracks to complications and recovery.
The plan · 6 months
From first paper to submission.
Project V follows the shared six-month rhythm below, and can continue afterward if the project leader proposes it.
Onboard
Read 10–15 key papers, set up data access and compute, write a one-page proposal.
Reproduce
Match a published baseline. Share a reproducibility report.
Propose
First version of the method, defended in a mentor review.
Iterate
Ablations, failure analysis, fix what the review found.
Validate
External data or pilot study; freeze results.
Submit
Paper, code release and a talk for the AIMA community.
Project V · Surgical AI
Want to work on
this question?
Apply through the form by November 7, 2026, choosing Project V, with an academic CV and a statement of purpose of up to 300 words. Interviews run November 9 to 13, and groups start around November 18. Mentors from Stanford University · Harvard University · VinUniversity · North Carolina A&T State University.
References
- Birkmeyer et al. Surgical Skill and Complication Rates after Bariatric Surgery. NEJM 2013. DOI
- Lingard. Communication failures in the operating room: an observational classification of recurrent types and effects. Quality & Safety in Health Care 2004. DOI
- Jung et al. First-year Analysis of the Operating Room Black Box Study. Annals of Surgery 2020. DOI
- Hatib et al. Machine-learning Algorithm to Predict Hypotension Based on High-fidelity Arterial Pressure Waveform Analysis. Anesthesiology 2018. DOI
- Wijnberge et al. Effect of a Machine Learning–Derived Early Warning System for Intraoperative Hypotension vs Standard Care on Depth and Duration of Intraoperative Hypotension During Elective Noncardiac Surgery. JAMA 2020. DOI
- Lambden et al. Definitions and pathophysiology of vasoplegic shock. Critical Care 2018. DOI
- Jung et al. Nontechnical Skill Assessment of the Collective Surgical Team Using the Non-Technical Skills for Surgeons (NOTSS) System. Annals of Surgery 2020. DOI
- Lee et al. VitalDB, a high-fidelity multi-parameter vital signs database in surgical patients. Scientific Data 2022. DOI
- Twinanda et al. EndoNet: A Deep Architecture for Recognition Tasks on Laparoscopic Videos. IEEE TMI 2017. DOI
- Özsoy et al. MM-OR: A Large Multimodal Operating Room Dataset for Semantic Understanding of High-Intensity Surgical Environments. CVPR 2025 · arXiv. arXiv
