Programs
Patient safety
Patient safety in care settings: early warning of falls, cross-checks on dispensing and prescription rules, and counting of surgical instruments.
Projects
Medication Administration and Dispensing Safety
Evaluate pill identification and prescription checks independently, then verify the evidence behind each error flag.
Fall Prediction and Prevention
Replay acceleration and gait records to measure warning lead time and distinguish falls from everyday movement.
Surgical Environment Error Prevention
Replay instrument labels and surgical phases to evaluate instrument state tracking and alert conditions.
Target, evidence, next check — kept apart
A target is what the project set out to reach. Evidence is what has been measured, in which scope. The next check is what would change the reading.
| Project | Internal target | Evidence so far | Next check |
|---|---|---|---|
| 18-2 · Medication safety | Zero missed contraindications or look-alike medication errors | Synthetic harness (dispense18) · Common synthetic checks | Train the vision modules on the fixed ePillID split, build per-category calibration sets from its validation split and fill the same table on its test split; categories below the minimum calibration size are reported as uncertified with no automatic pass. |
| 18-1 · Fall early warning | At least 92% early-warning sensitivity in the high-risk group | Common synthetic checks | Register participant and event splits, alert lead time and the time denominator for false alarms. |
| 18-3 · Surgical safety | Zero instrument count omissions on the fixed test | Common synthetic checks | Lock visibility, occlusion and tracking ID criteria and define separate video and synthetic count metrics. |
Evaluation data
Official ePillID consumer images by product type
MFDS pill identification and DUR with MIMIC-IV 3.1 as a candidate
UP-Fall Detection Dataset 2019
KFall 2021
FARSEEING observed falls and eICU-CRD 2.0
CAMMA Cholec80 videos and tool annotations
CholecTrack20 CVPR 2025
Independent instrument and sponge counts linked to video
Boundary cases recorded before evaluation
- Look-alike products under glare, distorted lighting and overlap
- Alerts for sound-alike products and unit mismatches
- Abstention when required lab values are missing or conflicting
- Event-level assessment of a warning at least three seconds before impact
- False alarms per hour from movements such as picking up objects or turning in bed
- Abstention with sensor displacement, device removal or missing samples
- Tracking and duplicate counts after smoke, occlusion or departure from the camera field
- A retained-item mismatch flagged before closure in synthetic sequences
- Single-frame latency of at most 33 ms on fixed hardware
What the evidence covers
Report a zero miss rate with the frozen test denominator. Product codes and source rules require verification across medical settings.
A laboratory fall label and an advance alert for a patient are different endpoints. Measure lead time only for records that cover at least three seconds before impact.
In-frame presence and segmentation cover a different observation space from total instrument inventories. Report the synthetic missed-count denominator.