Machine Learning Enabled Hind Foot Deformity Detection Using Individually Addressable Hybrid Pressure Sensor Matrix

We demonstrate a scalable, cost-effective hybrid sensor based on piezoelectric-triboelectric mechanisms for pressure mapping of 36 regions of the hindfoot and precise detection of any deformity. The matrix sensor comprises of individually addressable 6×6 sensors, which prevent pressure cell crosstalk and enhance the detection limit of the sensor up to 170kPa. The sensor can be easily integrated as an in-sole device because it is constructed on a flexible substrate using photo-patternable BTO (Barium Titanate)/SU-8 nanocomposite. The developed sensor has a pressure sensitivity of 34mV/kPa for 0-5kPa and 2.7mV/kPa for 5-170kPa, respectively. Additionally, we apply a convolution neural network (CNN-2D) to extract features from the 36 hybrid signals for precise deformity detection. The developed CNN model has a model loss of 0.1%, 98% accuracy, and a cross-feature differentiation value of 2%.