An Artificial Tactile Neuron Enabling Spiking Representation of Stiffness and Disease Diagnosis. Issue 24 (11th May 2022)

• Record Type: Journal Article
• Authors: Lee, Junseok; Kim, Seonjeong; Park, Seongjin; Lee, Jaesang; Hwang, Wonseop; Cho, Seong Won; Lee, Kyuho; Kim, Sun Mi; Seong, Tae‐Yeon; Park, Cheolmin; Lee, Suyoun; Yi, Hyunjung
• Is Part Of: Advanced materials. Volume 34:Issue 24(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-05-11
• Subjects: artificial tactile neurons -- disease diagnosis -- elastography -- neuromorphic sensors -- ovonic threshold switching -- piezoresistive sensors -- spiking neural networks; Materials -- Periodicals; Chemical vapor deposition -- Periodicals; 620.11
• DOI: 10.1002/adma.202201608
• Languages: English
• ISSNs: 0935-9648

Abstract

Abstract: Mechanical properties of biological systems provide useful information about the biochemical status of cells and tissues. Here, an artificial tactile neuron enabling spiking representation of stiffness and spiking neural network (SNN)‐based learning for disease diagnosis is reported. An artificial spiking tactile neuron based on an ovonic threshold switch serving as an artificial soma and a piezoresistive sensor as an artificial mechanoreceptor is developed and shown to encode the elastic stiffness of pressed materials into spike frequency evolution patterns. SNN‐based learning of ultrasound elastography images abstracted by spike frequency evolution rate enables the classification of malignancy status of breast tumors with a recognition accuracy up to 95.8%. The stiffness‐encoding artificial tactile neuron and learning of spiking‐represented stiffness patterns hold a great promise for the identification and classification of tumors for disease diagnosis and robot‐assisted surgery with low power consumption, low latency, and yet high accuracy. Abstract : An artificial tactile neuron that encodes the stiffness of pressed materials into spike frequency evolution patterns is developed using an ovonic threshold switch and a piezoresistive sensor. The spiking‐represented stiffness of soft materials with varying stiffness in a combination of spiking neural network‐based learning enables the classification of malignancy status of breast tumors with a recognition accuracy up to 95.8%.