Journal Paper In Progress

Prosthetic Knee Intelligence System (PKIS)

Real-Time Limb-Socket Monitoring via Stretchable E-Skin and Deep Learning

Over 60% of lower-limb prosthetic users develop skin complications in their first year of use — driven by invisible pressure patterns that static fitting cannot detect. PKIS transforms the prosthetic socket from a passive mechanical interface into a continuous, intelligent monitoring platform.

94.2%

HAR Accuracy

91.8%

Macro-F1

84.5%

Top-3 Forecast

92%

High-Risk Recall

28-40ms

CPU Latency

Project Overview

The Problem

Lower-limb amputees lack sensory feedback at the prosthetic socket interface. Skin-related complications exceed 60% in the first year, caused by non-uniform, time-varying pressure patterns invisible to static fitting protocols.

The Innovation

A stretchable piezoresistive e-skin (EGaIn/SiO₂/PDMS, GF≈24) embedded in the prosthetic liner, coupled with a CNN-BiLSTM-Attention multi-task deep learning framework for simultaneous activity recognition, forecasting, and risk assessment.

Key Design Choices

16-channel spatial pressure array · Dual-Arduino Bluetooth acquisition at 100 Hz · Safety-asymmetric loss weighting (λ_risk = 1.5) · Synthetic-first training with real-data fine-tuning.

Clinical Impact

First end-to-end soft-sensor-to-clinical-dashboard system for prosthetic interface monitoring. Enables proactive intervention before skin breakdown occurs, with real-time attention heatmaps for clinical interpretability.

PKIS signal acquisition pipeline: 16-channel E-Skin sensor array → dual-Arduino BLE acquisition at 100 Hz → signal conditioning → CNN-BiLSTM-Attention model → clinical dashboard — **Fig. 1.** End-to-end PKIS pipeline: stretchable e-skin sensor array → wireless BLE acquisition → signal conditioning → multi-task deep learning → real-time clinical dashboard.

Technical Details

Collaborators

Quorum Prosthetics · RockyTech · Xiao Lab, CU Boulder

Patient testing conducted with transtibial and transfemoral prosthetic users (IRB-approved, all permissions obtained).