Robotic Palpation of Fractures Using Bioinspired Tactile Sensor and Neuromorphic Encoding Algorithm

Palpation is a relatively safe, rapid, and low-cost method used by clinicians for examining diseased tissues. However, depending on the scanning speed and the physician’s experience, the size of the physical features in the body can be miscategorized or overlooked entirely. By designing tactile sensors and signal processing algorithms that mimic the body’s ability to account for variations in speed when scanning an object, we can solve the problem described above in an artificial system. We utilized a piezoresistive tactile sensor attached to a robotic arm to palpate fractures at different speeds. The analog tactile signals generated from the tactile sensor are converted into spike trains which are then scaled in time to encode the sensor data invariant of the speed of palpation. With a few principal components, the scaled dataset achieves a higher classification accuracy compared to the original dataset. Additionally, the scaled data was more robust to both spike timing noise and untrained speed conditions compared to the original data. Lastly, we demonstrated that this system could be applied in a medical setting by discriminating between 3 different fracture conditions (none, transverse, and communicated) in the ulna of a chicken wing with 99.8% accuracy at 3 different speeds.