Ulnar collateral ligament (UCL) injuries are a rising challenge among overhead athletes baseball pitchers, javelin throwers, and tennis players. UCL reconstruction (Tommy John surgery) increased 40% from 2010 to 2019 in MLB alone, with 54% of all reported injuries occurring in athletes aged 15–19. In New York State, reconstruction procedures increased 343% between 2003 and 2014, with 88.5% occurring in athletes aged 15–24.
For MLB pitchers, returning to game play after Tommy John surgery requires an average of 20.5 months (~600 days). Even then, performance is frequently reduced. Across the US, approximately 40,000 UCL-related injuries annually could immediately benefit from better rehabilitation monitoring tools.
Current tools for characterizing UCL stress during rehabilitation fall into three categories all with significant limitations:
Consumer wearables (e.g., Driveline's Pulse) are accessible but biomechanically inaccurate, carrying an estimated 38.7% error in elbow torque estimation.
MRI consultations can diagnose structural damage but cannot continuously assess dynamic elbow loading during throwing and rehabilitation activities.
Motion capture studios (e.g., Kinitrax Hawkeye) offer research-grade accuracy but cost over $1,000,000 to install and require controlled lab environments, making field deployment impossible.
No device on the market delivers accurate, personalized, non-invasive, and low-cost UCL stress analysis for athletes in the field.
As Team Captain and Project Manager for the Cornell DEBUT team, I led a group of eight engineers across all phases of development from initial problem scoping and stakeholder interviews (10+ clinical staff and D1 athletes) through to hardware iteration, software development, validation testing, and the VentureWell Phase 2B submission.
My direct contributions included leading the translation of the Hill's Musculoskeletal Model into the Python-based processing pipeline, coordinating the OptiTrack motion capture validation sessions at Cornell, overseeing PCB design reviews, and authoring the regulatory and commercialization sections of the feasibility report. I also led our market research immersion sessions with baseball and tennis players that informed the final wearability-versus-accuracy design tradeoffs.
Axio Recovery estimates valgus elbow torque using a patient-specific Hill's Musculoskeletal Model (HMM), calibrated to the individual athlete with a one-time setup protocol guided by the prescribing physician. The core equation is:
τ(θ, t) = Σ r(θ, t) · F(θ, t)
Where the radius component r represents the distance from each muscle head to the elbow joint (measured via IMUs), and the force component F represents the contraction of each target muscle group (measured via sEMG). The model targets the biceps brachii long head, triceps brachii long head, and pronator teres identified through extensive experimentation as dominant contributors (>10% each) to UCL stress.
Unlike generalized biomechanical models used in lab settings, the HMM is calibrated to each athlete's individual anthropometrics and muscle activation profiles, reducing reliance on population-average assumptions.
The physical system consists of wearable sensor bands and a waist-worn datalogger. MyoWare 2.0 sEMG sensors and LGD320 9DOF IMUs are encased in ABS plastic housings mounted on nylon compression bands. These connect to a Sparkfun ESP32-WROOM microcontroller via I2C (IMUs) and analog channels (sEMGs).
Custom PCBs include analog bandpass filters, USB-C charging, and I2C multiplexers. The ESP32 firmware (C++) handles signal processing at ~10 kHz sampling, compiling, and packing data to a microSD card. The system provides 3+ hours of continuous datalogging covering a full training session with a sensor band mass under 75g and surface temperature below 40°C.
The processing pipeline is built in Python and hosted on the Axio Recovery web platform. After a session, the athlete or clinician uploads the recorded .CSV from the datalogger. The Hill-Musculoskeletal pipeline processes the raw sEMG and IMU signals, applies the calibrated model, and outputs elbow torque estimates with session-level analytics for the physician.
The software stack is stored in an ISO-compliant GitLab repository. Edge case assessments including biceps curls and lateral arm movements confirmed motion-detection error below 5% for arm kinematics outside of throwing activities.
| Sensors | MyoWare 2.0 sEMG · LGD320 9DOF IMU |
| Microcontroller | Sparkfun ESP32-WROOM |
| Data Rate | ~10 kHz sampling |
| Battery Life | 3+ hours continuous logging |
| Band Mass | < 75 g per band |
| Surface Temp | < 40°C |
| Connectivity | MicroSD (current) · LoRa / BLE (planned) |
| Software Stack | Python · ISO-compliant GitLab · Cloud-hosted |
| Calibration | One-time physician-guided protocol |
| Target Muscles | Biceps brachii long head · Triceps brachii long head · Pronator teres |
| Regulatory Class | FDA Class II 21 CFR 890.1375 |
| Predicate Device | CMAP Pro™ (K113074) |
| Retail Price | $549.99 (3 bands + datalogger + software) |
| COGS | ~$353.73 per unit |
Verification testing with six participants confirmed the prototype satisfies all user-defined design requirements. Participants completed full baseball pitching regimens without observable restriction in range of motion, and no electrical dropouts or sensor migration occurred across an 8-hour continuous wear trial. One participant reported mild contact dermatitis from EKG pad exposure, flagged for resolution in the next iteration.
Preliminary validation was conducted in the OptiTrack motion capture studio at Cornell University. Low-intensity overhand throws (~20 mph) were recorded simultaneously with Axio Recovery and the OptiTrack system. Elbow torque from OptiTrack was computed via the Upper Extremity Dynamic Model in OpenSim with MATLAB post-processing using athlete anthropometrics.
In one representative trial, Axio Recovery estimated a peak torque of 55.6 N·m against the OptiTrack ground truth of 54.2 N·m yielding 2.58% error. This is a 15× improvement over the 38.7% error reported for Driveline's Pulse.
Motion Capture Studio Validation Session
The team submitted Axio Recovery as a Phase 2B entry to the VentureWell DEBUT (Design by Biomedical Undergraduate Teams) competition in June 2026. The submission covers the full commercialization pathway regulatory strategy (510(k) via 21 CFR 890.1375), reimbursement (RTM CPT codes 98975–98985), manufacturing cost analysis (COGS ~$353.73, retail $549.99), and a market impact analysis projecting a serviceable obtainable market of ~$7.22M annually at 5% adoption of the U.S. wearable sports monitoring segment.
DEBUT VentureWell Competition Submission Video
The immediate priority is wireless data streaming transitioning from wired Molex connectors to Low Range Radio (LoRa) or Bluetooth Low Energy (BLE) to improve athlete comfort. The next iteration also integrates a 2.7" LCD touchscreen into the datalogger for a digitized calibration walkthrough, replacing the current paper IFU-guided protocol.
A Comparative Effectiveness Research (CER) study comparing Axio Recovery against professional motion capture is in process for IRB approval through Cornell University. The study will recruit experienced baseball players and directly compare estimated valgus elbow torque against OptiTrack inverse kinematics across realistic throwing intensities establishing the evidence base for 510(k) submission and clinical adoption.