Project: Pharyvac FESS Computer Vision Date: 2026-03-28 Data: 16 bilateral full primary FESS cases (Nov 2025, Mar 2026)
The Core Idea
Every second in the OR costs money. By using computer vision to quantify exactly how surgical time is spent, instrument by instrument, swap by swap, pharyvac can identify where time is lost and what efficiency gains are possible.
OR Cost Context
From the spreadsheet’s reference data ($$$ sheet), operating room costs vary dramatically by facility:
| Facility Type | Per Minute | Per Hour |
|---|---|---|
| Academic medical centers (Columbia, NYU, Mount Sinai) | $60–100+ | $3,600–6,000+ |
| Community hospitals | $30–60 | $1,800–3,600 |
| Ambulatory surgery centers (ASCs) | $20–40 | $1,200–2,400 |
These costs include staff salaries (surgeons, anesthesiologists, nurses, techs), equipment depreciation, supplies, facility overhead, and malpractice insurance.
Where Time Goes: The Surgical Time Budget
Across 16 cases, here’s how the average case breaks down:
Mean recorded time: 2h 43min
| Category | Mean Time | % of Recorded | Cost @ $60/min | Cost @ $100/min |
|---|---|---|---|---|
| Forceps | 42:32 | 26.8% | $2,552 | $4,253 |
| Nav Suction | 22:49 | 13.2% | $1,369 | $2,282 |
| Microdebrider | 20:26 | 12.8% | $1,226 | $2,043 |
| Non-Nav Suction | 15:47 | 9.5% | $947 | $1,578 |
| Non-forcep Instruments | 16:21 | 9.6% | $981 | $1,635 |
| Nav Probe | 5:46 | 3.8% | $346 | $577 |
| Suction Bovie | 3:24 | 2.1% | $204 | $340 |
| Unlabelled/Transition | 40:46 | 25.4% | $2,446 | $4,077 |
| Unaccounted (non-instrument) | ~36 min | ~22% | $2,160 | $3,600 |
Total estimated OR cost per case: 8,100–16,300 (recorded time only, at academic center rates)
Instrument Swaps: Raw Counts vs True Physical Swaps
Each instrument swap involves the surgeon putting down one tool, potentially repositioning, and picking up another. This data captures something rarely quantified in surgical literature, but the raw YOLO-derived counts need careful interpretation.
Raw Data
| Metric | Value |
|---|---|
| Mean raw swaps per case | 434 |
| Raw swap rate | 2.7 swaps/minute |
| Range | 253, 638 |
| Correlation with case length | r = 0.89 |
The Inflation Problem
The raw count of 434 is likely inflated 2-4x by detection artifacts, primarily YOLO flickering between visually similar instrument classes frame-to-frame (especially among suction-type instruments). The evidence:
- At 434 swaps, the average instrument bout is only 16.8 seconds, unrealistically short for FESS
- Swap rate has suspiciously low variance (CV = 10.7%) across all cases regardless of complexity
- Swaps correlate r = 0.83 with nav suction time, suggesting class-boundary flickering between similar instruments
- Simple and complex cases have identical swap rates (2.7/min), which shouldn’t be the case for a real behavioral metric
See Swap Count Artifact Analysis for full details.
Corrected Estimates
After applying temporal smoothing (minimum 1-second bout duration), the estimated true physical swap count is 150-200 per case, with average bout durations of 36-49 seconds, much more consistent with FESS surgical workflow.
| Metric | Raw (Inflated) | Estimated True |
|---|---|---|
| Swaps per case | 434 | 150-200 |
| Avg bout duration | 16.8 sec | 36-49 sec |
| Swap rate | 2.7/min | ~1.0-1.3/min |
What Each True Swap Costs
Using the corrected estimate of 150-200 true swaps at 3-5 seconds per transition:
| Swap Duration | True Swaps | Time Lost Per Case | Cost @ $60/min | Cost @ $100/min |
|---|---|---|---|---|
| 3 seconds | 175 | 8.8 min | $525 | $875 |
| 5 seconds | 175 | 14.6 min | $875 | $1,458 |
Corrected estimate: instrument transitions consume ~525–1,460 per case at an academic medical center. While lower than the original estimate, over hundreds of FESS cases per year at a busy center, this still adds up to meaningful savings. The total unaccounted time (~22% of recorded time, measured directly from video) remains the more reliable efficiency metric.
Simple vs Complex Cases: Cost Implications
| Metric | Simple FESS (n=7) | Complex FESS (n=9) | Delta |
|---|---|---|---|
| Mean case time | 2:04:25 | 3:09:46 | +1:05:21 |
| Mean raw swaps | 337 | 510 | +173 |
| Raw swap rate (per min) | 2.7 | 2.7 | ~same (artifact) |
| Est. cost @ $60/min | $7,465 | $11,387 | +$3,922 |
| Est. cost @ $100/min | $12,442 | $18,978 | +$6,537 |
The identical swap rate across complexity levels (2.7/min for both) is itself evidence of the detection artifact, a real behavioral metric should differ between simple and complex cases. After smoothing, the swap rate difference between simple and complex cases will be a more meaningful comparison.
Actionable Efficiency Targets
1. Quantify True Transition Time
Before estimating savings from reducing swaps, implement temporal smoothing to get accurate swap counts. The unaccounted time metric (~22% of recorded time, ~36 min/case) is currently the most reliable measure of non-productive OR time. At academic center rates, this represents 2,160–3,600 per case in time where no instrument is actively in use.
2. Reduce True Instrument Swaps by 10%
Once smoothed swap counts are available, if ergonomic tray redesign or instrument sequencing could eliminate 10% of true swaps:
- Saves: ~15-20 true swaps → ~1-2 min per case
- Annual savings (200 cases/yr): 3-7 hours of OR time →
12,000–40,000
3. Reduce Unlabelled/Transition Time by 25%
Better instrument handoff workflows, pre-staged instruments, or surgical tech anticipation:
- Saves: ~10 min per case
- Annual savings (200 cases/yr): 33 hours →
60,000–200,000
4. Standardize Instrument Sequencing
The high variance in instrument time proportions (e.g., forceps ranges from 20.9% to 39.0%) suggests significant surgeon-to-surgeon or case-to-case variation. These proportional measures are reliable (unaffected by detection flickering) and can be used to identify the most efficient instrument sequencing patterns from top-performing cases.
Making the Business Case
For Hospital Administrators
The data shows a quantifiable, objective measure of OR time utilization at the instrument level, something no existing system provides. The value proposition:
- Real-time OR efficiency dashboards
- Surgeon-specific benchmarking (anonymized)
- Identification of bottlenecks that are invisible to the naked eye
- ROI: even a 5% reduction in average case time pays for the system many times over
For Surgeons
- Objective feedback on operative efficiency without subjective assessment
- Identify personal patterns (e.g., “you use 15% more forceps time than the cohort average”)
- Training tool for residents: compare trainee instrument patterns to attending benchmarks
For Payers/Insurers
- Evidence-based surgical efficiency metrics
- Potential for value-based reimbursement tied to measurable efficiency outcomes
Key Numbers for Presentations
~175 true instrument swaps per case, each one a moment the surgeon isn’t operating (raw YOLO count of 434 is inflated by detection flickering; see Swap Count Artifact Analysis)
525–1,460 per case in estimated transition time costs (corrected from detection artifacts)
22% of OR time is non-instrument “dead time”, directly measured, the most reliable efficiency metric
25% of OR time is currently unclassified by the CV model, a data gap that improving the model can close
Instrument time proportions are reliable, forceps 26.8%, nav suction 13.2%, microdebrider 12.8%, unaffected by detection artifacts